An extensive benchmark study on biomedical text generation and mining with ChatGPT

Evaluating and mitigating bias in AI-based medical text generation

Artificial intelligence (AI) systems, particularly those based on deep learning models, have increasingly achieved expert-level performance in medical applications. However, there is growing concern that such AI systems may reflect and amplify human bias, reducing the quality of their performance in historically underserved populations. The fairness issue has attracted considerable research interest in the medical imaging classification field, yet it remains understudied in the text-generation domain. In this study, we investigate the fairness problem in text generation within the medical field and observe substantial performance discrepancies across different races, sexes and age groups, including intersectional groups, various model scales and different evaluation metrics. To mitigate this fairness issue, we propose an algorithm that selectively optimizes those underserved groups to reduce bias. Our evaluations across multiple backbones, datasets and modalities demonstrate that our proposed algorithm enhances fairness in text generation without compromising overall performance.

Unravelling Convergent Signaling Mechanisms Underlying the Aging-Disease Nexus Using Computational Language Analysis

Multiple lines of evidence suggest that multiple pathological conditions and diseases that account for the majority of human mortality are driven by the molecular aging process. At the cellular level, aging can largely be conceptualized to comprise the progressive accumulation of molecular damage, leading to resultant cellular dysfunction. As many diseases, e.g., cancer, coronary heart disease, Chronic obstructive pulmonary disease, Type II diabetes mellitus, or chronic kidney disease, potentially share a common molecular etiology, then the identification of such mechanisms may represent an ideal locus to develop targeted prophylactic agents that can mitigate this disease-driving mechanism. Here, using the input of artificial intelligence systems to generate unbiased disease and aging mechanism profiles, we have aimed to identify key signaling mechanisms that may represent new disease-preventing signaling pathways that are ideal for the creation of disease-preventing chemical interventions. Using a combinatorial informatics approach, we have identified a potential critical mechanism involving the recently identified kinase, Dual specificity tyrosine-phosphorylation-regulated kinase 3 (DYRK3) and the epidermal growth factor receptor (EGFR) that may function as a regulator of the pathological transition of health into disease via the control of cellular fate in response to stressful insults.

Comparative analysis of generative LLMs for labeling entities in clinical notes

This paper evaluates and compares different fine-tuned variations of generative large language models (LLM) in the zero-shot named entity recognition (NER) task for the clinical domain. As part of the 8th Biomedical Linked Annotation Hackathon, we examined Llama 2 and Mistral models, including base versions and those that have been fine-tuned for code, chat, and instruction-following tasks. We assess both the number of correctly identified entities and the models' ability to retrieve entities in structured formats. We used a publicly available set of clinical cases labeled with mentions of diseases, symptoms, and medical procedures for the evaluation. Results show that instruction fine-tuned models perform better than chat fine-tuned and base models in recognizing entities. It is also shown that models perform better when simple output structures are requested.

The Potential of Gemini and GPTs for Structured Report Generation based on Free-Text 18F-FDG PET/CT Breast Cancer Reports

RATIONALE AND OBJECTIVE

To compare the performance of large language model (LLM) based Gemini and Generative Pre-trained Transformers (GPTs) in data mining and generating structured reports based on free-text PET/CT reports for breast cancer after user-defined tasks.

MATERIALS AND METHODS

Breast cancer patients (mean age, 50 years ± 11 [SD]; all female) who underwent consecutive 18F-FDG PET/CT for follow-up between July 2005 and October 2023 were retrospectively included in the study. A total of twenty reports from 10 patients were used to train user-defined text prompts for Gemini and GPTs, by which structured PET/CT reports were generated. The natural language processing (NLP) generated structured reports and the structured reports annotated by nuclear medicine physicians were compared in terms of data extraction accuracy and capacity of progress decision-making. Statistical methods, including chi-square test, McNemar test and paired samples t-test, were employed in the study.

RESULTS

The structured PET/CT reports for 131 patients were generated by using the two NLP techniques, including Gemini and GPTs. In general, GPTs exhibited superiority over Gemini in data mining in terms of primary lesion size (89.6% vs. 53.8%, p < 0.001) and metastatic lesions (96.3% vs 89.6%, p < 0.001). Moreover, GPTs outperformed Gemini in making decision for progress (p < 0.001) and semantic similarity (F1 score 0.930 vs 0.907, p < 0.001) for reports.

CONCLUSION

GPTs outperformed Gemini in generating structured reports based on free-text PET/CT reports, which is potentially applied in clinical practice.

DATA AVAILABILITY

The data used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Aligning Large Language Models with Humans: A Comprehensive Survey of ChatGPT's Aptitude in Pharmacology

BACKGROUND

Due to the lack of a comprehensive pharmacology test set, evaluating the potential and value of large language models (LLMs) in pharmacology is complex and challenging.

AIMS

This study aims to provide a test set reference for assessing the application potential of both general-purpose and specialized LLMs in pharmacology.

METHODS

We constructed a pharmacology test set consisting of three tasks: drug information retrieval, lead compound structure optimization, and research trend summarization and analysis. Subsequently, we compared the performance of general-purpose LLMs GPT-3.5 and GPT-4 on this test set.

RESULTS

The results indicate that GPT-3.5 and GPT-4 can better understand instructions for information retrieval, scheme optimization, and trend summarization in pharmacology, showing significant potential in basic pharmacology tasks, especially in areas such as drug pharmacological properties, pharmacokinetics, mode of action, and toxicity prediction. These general LLMs also effectively summarize the current challenges and future trends in this field, proving their valuable resource for interdisciplinary pharmacology researchers. However, the limitations of ChatGPT become evident when handling tasks such as drug identification queries, drug interaction information retrieval, and drug structure simulation optimization. It struggles to provide accurate interaction information for individual or specific drugs and cannot optimize specific drugs. This lack of depth in knowledge integration and analysis limits its application in scientific research and clinical exploration.

CONCLUSION

Therefore, exploring retrieval-augmented generation (RAG) or integrating proprietary knowledge bases and knowledge graphs into pharmacology-oriented ChatGPT systems would yield favorable results. This integration will further optimize the potential of LLMs in pharmacology.

Testing and Evaluation of Health Care Applications of Large Language Models: A Systematic Review

Importance

Large language models (LLMs) can assist in various health care activities, but current evaluation approaches may not adequately identify the most useful application areas.

Objective

To summarize existing evaluations of LLMs in health care in terms of 5 components: (1) evaluation data type, (2) health care task, (3) natural language processing (NLP) and natural language understanding (NLU) tasks, (4) dimension of evaluation, and (5) medical specialty.

Data Sources

A systematic search of PubMed and Web of Science was performed for studies published between January 1, 2022, and February 19, 2024.

Study Selection

Studies evaluating 1 or more LLMs in health care.

Data Extraction and Synthesis

Three independent reviewers categorized studies via keyword searches based on the data used, the health care tasks, the NLP and NLU tasks, the dimensions of evaluation, and the medical specialty.

Results

Of 519 studies reviewed, published between January 1, 2022, and February 19, 2024, only 5% used real patient care data for LLM evaluation. The most common health care tasks were assessing medical knowledge such as answering medical licensing examination questions (44.5%) and making diagnoses (19.5%). Administrative tasks such as assigning billing codes (0.2%) and writing prescriptions (0.2%) were less studied. For NLP and NLU tasks, most studies focused on question answering (84.2%), while tasks such as summarization (8.9%) and conversational dialogue (3.3%) were infrequent. Almost all studies (95.4%) used accuracy as the primary dimension of evaluation; fairness, bias, and toxicity (15.8%), deployment considerations (4.6%), and calibration and uncertainty (1.2%) were infrequently measured. Finally, in terms of medical specialty area, most studies were in generic health care applications (25.6%), internal medicine (16.4%), surgery (11.4%), and ophthalmology (6.9%), with nuclear medicine (0.6%), physical medicine (0.4%), and medical genetics (0.2%) being the least represented.

Conclusions and Relevance

Existing evaluations of LLMs mostly focus on accuracy of question answering for medical examinations, without consideration of real patient care data. Dimensions such as fairness, bias, and toxicity and deployment considerations received limited attention. Future evaluations should adopt standardized applications and metrics, use clinical data, and broaden focus to include a wider range of tasks and specialties.

A multimodal approach for few-shot biomedical named entity recognition in low-resource languages

In this study, we revisit named entity recognition (NER) in the biomedical domain from a multimodal perspective, with a particular focus on applications in low-resource languages. Existing research primarily relies on unimodal methods for NER, which limits the potential for capturing diverse information. To address this limitation, we propose a novel method that integrates a cross-modal generation module to transform unimodal data into multimodal data, thereby enabling the use of enriched multimodal information for NER. Additionally, we design a cross-modal filtering module to mitigate the adverse effects of text-image mismatches in multimodal NER. We validate our proposed method on two biomedical datasets specifically curated for low-resource languages. Experimental results demonstrate that our method significantly enhances the performance of NER, highlighting its effectiveness and potential for broader applications in biomedical research and low-resource language contexts.

FuncFetch: an LLM-assisted workflow enables mining thousands of enzyme-substrate interactions from published manuscripts

MOTIVATION

Thousands of genomes are publicly available, however, most genes in those genomes have poorly defined functions. This is partly due to a gap between previously published, experimentally characterized protein activities and activities deposited in databases. This activity deposition is bottlenecked by the time-consuming biocuration process. The emergence of large language models presents an opportunity to speed up the text-mining of protein activities for biocuration.

RESULTS

We developed FuncFetch-a workflow that integrates NCBI E-Utilities, OpenAI's GPT-4, and Zotero-to screen thousands of manuscripts and extract enzyme activities. Extensive validation revealed high precision and recall of GPT-4 in determining whether the abstract of a given paper indicates the presence of a characterized enzyme activity in that paper. Provided the manuscript, FuncFetch extracted data such as species information, enzyme names, sequence identifiers, substrates, and products, which were subjected to extensive quality analyses. Comparison of this workflow against a manually curated dataset of BAHD acyltransferase activities demonstrated a precision/recall of 0.86/0.64 in extracting substrates. We further deployed FuncFetch on nine large plant enzyme families. Screening 26 543 papers, FuncFetch retrieved 32 605 entries from 5459 selected papers. We also identified multiple extraction errors including incorrect associations, nontarget enzymes, and hallucinations, which highlight the need for further manual curation. The BAHD activities were verified, resulting in a comprehensive functional fingerprint of this family and revealing that ∼70% of the experimentally characterized enzymes are uncurated in the public domain. FuncFetch represents an advance in biocuration and lays the groundwork for predicting the functions of uncharacterized enzymes.

AVAILABILITY AND IMPLEMENTATION

Code and minimally curated activities are available at: https://github.com/moghelab/funcfetch and https://tools.moghelab.org/funczymedb.

Advancing psychosocial disability and psychosocial rehabilitation research through large language models and computational text mining

Psychosocial rehabilitation and psychosocial disability research have been a longstanding topic in healthcare, demanding continuous exploration and analysis to enhance patient and clinical outcomes. As the prevalence of psychosocial disability research continues to attract scholarly attention, many scientific articles are being published in the literature. These publications offer profound insights into diagnostics, preventative measures, treatment strategies, and epidemiological factors. Computational text mining as a subfield of artificial intelligence (AI) can make a big difference in accurately analyzing the current extensive collection of scientific articles on time, assisting individual scientists in understanding psychosocial disabilities better, and improving how we care for people with these challenges. Leveraging the vast repository of scientific literature available on PubMed, this study employs advanced text mining strategies, including word embeddings and large language models (LLMs) to extract valuable insights, automatically catalyzing research in mental health. It aims to significantly enhance the scientific community's knowledge by creating an extensive textual dataset and advanced computational text mining strategies to explore current trends in psychosocial rehabilitation and psychosocial disability research.

Bioinformatics and biomedical informatics with ChatGPT: Year one review

The year 2023 marked a significant surge in the exploration of applying large language model chatbots, notably Chat Generative Pre-trained Transformer (ChatGPT), across various disciplines. We surveyed the application of ChatGPT in bioinformatics and biomedical informatics throughout the year, covering omics, genetics, biomedical text mining, drug discovery, biomedical image understanding, bioinformatics programming, and bioinformatics education. Our survey delineates the current strengths and limitations of this chatbot in bioinformatics and offers insights into potential avenues for future developments.

Utilizing active learning strategies in machine-assisted annotation for clinical named entity recognition: a comprehensive analysis considering annotation costs and target effectiveness

OBJECTIVES

Active learning (AL) has rarely integrated diversity-based and uncertainty-based strategies into a dynamic sampling framework for clinical named entity recognition (NER). Machine-assisted annotation is becoming popular for creating gold-standard labels. This study investigated the effectiveness of dynamic AL strategies under simulated machine-assisted annotation scenarios for clinical NER.

MATERIALS AND METHODS

We proposed 3 new AL strategies: a diversity-based strategy (CLUSTER) based on Sentence-BERT and 2 dynamic strategies (CLC and CNBSE) capable of switching from diversity-based to uncertainty-based strategies. Using BioClinicalBERT as the foundational NER model, we conducted simulation experiments on 3 medication-related clinical NER datasets independently: i2b2 2009, n2c2 2018 (Track 2), and MADE 1.0. We compared the proposed strategies with uncertainty-based (LC and NBSE) and passive-learning (RANDOM) strategies. Performance was primarily measured by the number of edits made by the annotators to achieve a desired target effectiveness evaluated on independent test sets.

RESULTS

When aiming for 98% overall target effectiveness, on average, CLUSTER required the fewest edits. When aiming for 99% overall target effectiveness, CNBSE required 20.4% fewer edits than NBSE did. CLUSTER and RANDOM could not achieve such a high target under the pool-based simulation experiment. For high-difficulty entities, CNBSE required 22.5% fewer edits than NBSE to achieve 99% target effectiveness, whereas neither CLUSTER nor RANDOM achieved 93% target effectiveness.

DISCUSSION AND CONCLUSION

When the target effectiveness was set high, the proposed dynamic strategy CNBSE exhibited both strong learning capabilities and low annotation costs in machine-assisted annotation. CLUSTER required the fewest edits when the target effectiveness was set low.

Assessing the utility of large language models for phenotype-driven gene prioritization in the diagnosis of rare genetic disease

Phenotype-driven gene prioritization is fundamental to diagnosing rare genetic disorders. While traditional approaches rely on curated knowledge graphs with phenotype-gene relations, recent advancements in large language models (LLMs) promise a streamlined text-to-gene solution. In this study, we evaluated five LLMs, including two generative pre-trained transformers (GPT) series and three Llama2 series, assessing their performance across task completeness, gene prediction accuracy, and adherence to required output structures. We conducted experiments, exploring various combinations of models, prompts, phenotypic input types, and task difficulty levels. Our findings revealed that the best-performed LLM, GPT-4, achieved an average accuracy of 17.0% in identifying diagnosed genes within the top 50 predictions, which still falls behind traditional tools. However, accuracy increased with the model size. Consistent results were observed over time, as shown in the dataset curated after 2023. Advanced techniques such as retrieval-augmented generation (RAG) and few-shot learning did not improve the accuracy. Sophisticated prompts were more likely to enhance task completeness, especially in smaller models. Conversely, complicated prompts tended to decrease output structure compliance rate. LLMs also achieved better-than-random prediction accuracy with free-text input, though performance was slightly lower than with standardized concept input. Bias analysis showed that highly cited genes, such as BRCA1, TP53, and PTEN, are more likely to be predicted. Our study provides valuable insights into integrating LLMs with genomic analysis, contributing to the ongoing discussion on their utilization in clinical workflows.

Prompt Engineering Paradigms for Medical Applications: Scoping Review

BACKGROUND

Prompt engineering, focusing on crafting effective prompts to large language models (LLMs), has garnered attention for its capabilities at harnessing the potential of LLMs. This is even more crucial in the medical domain due to its specialized terminology and language technicity. Clinical natural language processing applications must navigate complex language and ensure privacy compliance. Prompt engineering offers a novel approach by designing tailored prompts to guide models in exploiting clinically relevant information from complex medical texts. Despite its promise, the efficacy of prompt engineering in the medical domain remains to be fully explored.

OBJECTIVE

The aim of the study is to review research efforts and technical approaches in prompt engineering for medical applications as well as provide an overview of opportunities and challenges for clinical practice.

METHODS

Databases indexing the fields of medicine, computer science, and medical informatics were queried in order to identify relevant published papers. Since prompt engineering is an emerging field, preprint databases were also considered. Multiple data were extracted, such as the prompt paradigm, the involved LLMs, the languages of the study, the domain of the topic, the baselines, and several learning, design, and architecture strategies specific to prompt engineering. We include studies that apply prompt engineering-based methods to the medical domain, published between 2022 and 2024, and covering multiple prompt paradigms such as prompt learning (PL), prompt tuning (PT), and prompt design (PD).

RESULTS

We included 114 recent prompt engineering studies. Among the 3 prompt paradigms, we have observed that PD is the most prevalent (78 papers). In 12 papers, PD, PL, and PT terms were used interchangeably. While ChatGPT is the most commonly used LLM, we have identified 7 studies using this LLM on a sensitive clinical data set. Chain-of-thought, present in 17 studies, emerges as the most frequent PD technique. While PL and PT papers typically provide a baseline for evaluating prompt-based approaches, 61% (48/78) of the PD studies do not report any nonprompt-related baseline. Finally, we individually examine each of the key prompt engineering-specific information reported across papers and find that many studies neglect to explicitly mention them, posing a challenge for advancing prompt engineering research.

CONCLUSIONS

In addition to reporting on trends and the scientific landscape of prompt engineering, we provide reporting guidelines for future studies to help advance research in the medical field. We also disclose tables and figures summarizing medical prompt engineering papers available and hope that future contributions will leverage these existing works to better advance the field.

PharmaBench: Enhancing ADMET benchmarks with large language models

Accurately predicting ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) properties early in drug development is essential for selecting compounds with optimal pharmacokinetics and minimal toxicity. Existing ADMET-related benchmark sets are limited in utility due to their small dataset sizes and the lack of representation of compounds used in drug discovery projects. These shortcomings hinder their application in model building for drug discovery. To address this issue, we propose a multi-agent data mining system based on Large Language Models that effectively identifies experimental conditions within 14,401 bioassays. This approach facilitates merging entries from different sources, culminating in the creation of PharmaBench. Additionally, we have developed a data processing workflow to integrate data from various sources, resulting in 156,618 raw entries. Through this workflow, we constructed PharmaBench, a comprehensive benchmark set for ADMET properties, which comprises eleven ADMET datasets and 52,482 entries. This benchmark set is designed to serve as an open-source dataset for the development of AI models relevant to drug discovery projects.

Evaluating the Capabilities of Generative AI Tools in Understanding Medical Papers: Qualitative Study

BACKGROUND

Reading medical papers is a challenging and time-consuming task for doctors, especially when the papers are long and complex. A tool that can help doctors efficiently process and understand medical papers is needed.

OBJECTIVE

This study aims to critically assess and compare the comprehension capabilities of large language models (LLMs) in accurately and efficiently understanding medical research papers using the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) checklist, which provides a standardized framework for evaluating key elements of observational study.

METHODS

The study is a methodological type of research. The study aims to evaluate the understanding capabilities of new generative artificial intelligence tools in medical papers. A novel benchmark pipeline processed 50 medical research papers from PubMed, comparing the answers of 6 LLMs (GPT-3.5-Turbo, GPT-4-0613, GPT-4-1106, PaLM 2, Claude v1, and Gemini Pro) to the benchmark established by expert medical professors. Fifteen questions, derived from the STROBE checklist, assessed LLMs' understanding of different sections of a research paper.

RESULTS

LLMs exhibited varying performance, with GPT-3.5-Turbo achieving the highest percentage of correct answers (n=3916, 66.9%), followed by GPT-4-1106 (n=3837, 65.6%), PaLM 2 (n=3632, 62.1%), Claude v1 (n=2887, 58.3%), Gemini Pro (n=2878, 49.2%), and GPT-4-0613 (n=2580, 44.1%). Statistical analysis revealed statistically significant differences between LLMs (P<.001), with older models showing inconsistent performance compared to newer versions. LLMs showcased distinct performances for each question across different parts of a scholarly paper-with certain models like PaLM 2 and GPT-3.5 showing remarkable versatility and depth in understanding.

CONCLUSIONS

This study is the first to evaluate the performance of different LLMs in understanding medical papers using the retrieval augmented generation method. The findings highlight the potential of LLMs to enhance medical research by improving efficiency and facilitating evidence-based decision-making. Further research is needed to address limitations such as the influence of question formats, potential biases, and the rapid evolution of LLM models.

STRING-ing together protein complexes: corpus and methods for extracting physical protein interactions from the biomedical literature

MOTIVATION

Understanding biological processes relies heavily on curated knowledge of physical interactions between proteins. Yet, a notable gap remains between the information stored in databases of curated knowledge and the plethora of interactions documented in the scientific literature.

RESULTS

To bridge this gap, we introduce ComplexTome, a manually annotated corpus designed to facilitate the development of text-mining methods for the extraction of complex formation relationships among biomedical entities targeting the downstream semantics of the physical interaction subnetwork of the STRING database. This corpus comprises 1287 documents with ∼3500 relationships. We train a novel relation extraction model on this corpus and find that it can highly reliably identify physical protein interactions (F1-score = 82.8%). We additionally enhance the model's capabilities through unsupervised trigger word detection and apply it to extract relations and trigger words for these relations from all open publications in the domain literature. This information has been fully integrated into the latest version of the STRING database.

AVAILABILITY AND IMPLEMENTATION

We provide the corpus, code, and all results produced by the large-scale runs of our systems biomedical on literature via Zenodo https://doi.org/10.5281/zenodo.8139716, Github https://github.com/farmeh/ComplexTome_extraction, and the latest version of STRING database https://string-db.org/.

Taiyi: a bilingual fine-tuned large language model for diverse biomedical tasks

OBJECTIVE

Most existing fine-tuned biomedical large language models (LLMs) focus on enhancing performance in monolingual biomedical question answering and conversation tasks. To investigate the effectiveness of the fine-tuned LLMs on diverse biomedical natural language processing (NLP) tasks in different languages, we present Taiyi, a bilingual fine-tuned LLM for diverse biomedical NLP tasks.

MATERIALS AND METHODS

We first curated a comprehensive collection of 140 existing biomedical text mining datasets (102 English and 38 Chinese datasets) across over 10 task types. Subsequently, these corpora were converted to the instruction data used to fine-tune the general LLM. During the supervised fine-tuning phase, a 2-stage strategy is proposed to optimize the model performance across various tasks.

RESULTS

Experimental results on 13 test sets, which include named entity recognition, relation extraction, text classification, and question answering tasks, demonstrate that Taiyi achieves superior performance compared to general LLMs. The case study involving additional biomedical NLP tasks further shows Taiyi's considerable potential for bilingual biomedical multitasking.

CONCLUSION

Leveraging rich high-quality biomedical corpora and developing effective fine-tuning strategies can significantly improve the performance of LLMs within the biomedical domain. Taiyi shows the bilingual multitasking capability through supervised fine-tuning. However, those tasks such as information extraction that are not generation tasks in nature remain challenging for LLM-based generative approaches, and they still underperform the conventional discriminative approaches using smaller language models.

VAIV bio-discovery service using transformer model and retrieval augmented generation

BACKGROUND

There has been a considerable advancement in AI technologies like LLM and machine learning to support biomedical knowledge discovery.

MAIN BODY

We propose a novel biomedical neural search service called 'VAIV Bio-Discovery', which supports enhanced knowledge discovery and document search on unstructured text such as PubMed. It mainly handles with information related to chemical compound/drugs, gene/proteins, diseases, and their interactions (chemical compounds/drugs-proteins/gene including drugs-targets, drug-drug, and drug-disease). To provide comprehensive knowledge, the system offers four search options: basic search, entity and interaction search, and natural language search. We employ T5slim_dec, which adapts the autoregressive generation task of the T5 (text-to-text transfer transformer) to the interaction extraction task by removing the self-attention layer in the decoder block. It also assists in interpreting research findings by summarizing the retrieved search results for a given natural language query with Retrieval Augmented Generation (RAG). The search engine is built with a hybrid method that combines neural search with the probabilistic search, BM25.

CONCLUSION

As a result, our system can better understand the context, semantics and relationships between terms within the document, enhancing search accuracy. This research contributes to the rapidly evolving biomedical field by introducing a new service to access and discover relevant knowledge.

Comparison of Medical Research Abstracts Written by Surgical Trainees and Senior Surgeons or Generated by Large Language Models

Importance

Artificial intelligence (AI) has permeated academia, especially OpenAI Chat Generative Pretrained Transformer (ChatGPT), a large language model. However, little has been reported on its use in medical research.

Objective

To assess a chatbot's capability to generate and grade medical research abstracts.

Design, Setting, and Participants

In this cross-sectional study, ChatGPT versions 3.5 and 4.0 (referred to as chatbot 1 and chatbot 2) were coached to generate 10 abstracts by providing background literature, prompts, analyzed data for each topic, and 10 previously presented, unassociated abstracts to serve as models. The study was conducted between August 2023 and February 2024 (including data analysis).

Exposure

Abstract versions utilizing the same topic and data were written by a surgical trainee or a senior physician or generated by chatbot 1 and chatbot 2 for comparison. The 10 training abstracts were written by 8 surgical residents or fellows, edited by the same senior surgeon, at a high-volume hospital in the Southeastern US with an emphasis on outcomes-based research. Abstract comparison was then based on 10 abstracts written by 5 surgical trainees within the first 6 months of their research year, edited by the same senior author.

Main Outcomes and Measures

The primary outcome measurements were the abstract grades using 10- and 20-point scales and ranks (first to fourth). Abstract versions by chatbot 1, chatbot 2, junior residents, and the senior author were compared and judged by blinded surgeon-reviewers as well as both chatbot models. Five academic attending surgeons from Denmark, the UK, and the US, with extensive experience in surgical organizations, research, and abstract evaluation served as reviewers.

Results

Surgeon-reviewers were unable to differentiate between abstract versions. Each reviewer ranked an AI-generated version first at least once. Abstracts demonstrated no difference in their median (IQR) 10-point scores (resident, 7.0 [6.0-8.0]; senior author, 7.0 [6.0-8.0]; chatbot 1, 7.0 [6.0-8.0]; chatbot 2, 7.0 [6.0-8.0]; P = .61), 20-point scores (resident, 14.0 [12.0-7.0]; senior author, 15.0 [13.0-17.0]; chatbot 1, 14.0 [12.0-16.0]; chatbot 2, 14.0 [13.0-16.0]; P = .50), or rank (resident, 3.0 [1.0-4.0]; senior author, 2.0 [1.0-4.0]; chatbot 1, 3.0 [2.0-4.0]; chatbot 2, 2.0 [1.0-3.0]; P = .14). The abstract grades given by chatbot 1 were comparable to the surgeon-reviewers' grades. However, chatbot 2 graded more favorably than the surgeon-reviewers and chatbot 1. Median (IQR) chatbot 2-reviewer grades were higher than surgeon-reviewer grades of all 4 abstract versions (resident, 14.0 [12.0-17.0] vs 16.9 [16.0-17.5]; P = .02; senior author, 15.0 [13.0-17.0] vs 17.0 [16.5-18.0]; P = .03; chatbot 1, 14.0 [12.0-16.0] vs 17.8 [17.5-18.5]; P = .002; chatbot 2, 14.0 [13.0-16.0] vs 16.8 [14.5-18.0]; P = .04). When comparing the grades of the 2 chatbots, chatbot 2 gave higher median (IQR) grades for abstracts than chatbot 1 (resident, 14.0 [13.0-15.0] vs 16.9 [16.0-17.5]; P = .003; senior author, 13.5 [13.0-15.5] vs 17.0 [16.5-18.0]; P = .004; chatbot 1, 14.5 [13.0-15.0] vs 17.8 [17.5-18.5]; P = .003; chatbot 2, 14.0 [13.0-15.0] vs 16.8 [14.5-18.0]; P = .01).

Conclusions and Relevance

In this cross-sectional study, trained chatbots generated convincing medical abstracts, undifferentiable from resident or senior author drafts. Chatbot 1 graded abstracts similarly to surgeon-reviewers, while chatbot 2 was less stringent. These findings may assist surgeon-scientists in successfully implementing AI in medical research.

Fine-tuning large language models for chemical text mining

Extracting knowledge from complex and diverse chemical texts is a pivotal task for both experimental and computational chemists. The task is still considered to be extremely challenging due to the complexity of the chemical language and scientific literature. This study explored the power of fine-tuned large language models (LLMs) on five intricate chemical text mining tasks: compound entity recognition, reaction role labelling, metal-organic framework (MOF) synthesis information extraction, nuclear magnetic resonance spectroscopy (NMR) data extraction, and the conversion of reaction paragraphs to action sequences. The fine-tuned LLMs demonstrated impressive performance, significantly reducing the need for repetitive and extensive prompt engineering experiments. For comparison, we guided ChatGPT (GPT-3.5-turbo) and GPT-4 with prompt engineering and fine-tuned GPT-3.5-turbo as well as other open-source LLMs such as Mistral, Llama3, Llama2, T5, and BART. The results showed that the fine-tuned ChatGPT models excelled in all tasks. They achieved exact accuracy levels ranging from 69% to 95% on these tasks with minimal annotated data. They even outperformed those task-adaptive pre-training and fine-tuning models that were based on a significantly larger amount of in-domain data. Notably, fine-tuned Mistral and Llama3 show competitive abilities. Given their versatility, robustness, and low-code capability, leveraging fine-tuned LLMs as flexible and effective toolkits for automated data acquisition could revolutionize chemical knowledge extraction.

The changing landscape of text mining: a review of approaches for ecology and evolution

In ecology and evolutionary biology, the synthesis and modelling of data from published literature are commonly used to generate insights and test theories across systems. However, the tasks of searching, screening, and extracting data from literature are often arduous. Researchers may manually process hundreds to thousands of articles for systematic reviews, meta-analyses, and compiling synthetic datasets. As relevant articles expand to tens or hundreds of thousands, computer-based approaches can increase the efficiency, transparency and reproducibility of literature-based research. Methods available for text mining are rapidly changing owing to developments in machine learning-based language models. We review the growing landscape of approaches, mapping them onto three broad paradigms (frequency-based approaches, traditional Natural Language Processing and deep learning-based language models). This serves as an entry point to learn foundational and cutting-edge concepts, vocabularies, and methods to foster integration of these tools into ecological and evolutionary research. We cover approaches for modelling ecological texts, generating training data, developing custom models and interacting with large language models and discuss challenges and possible solutions to implementing these methods in ecology and evolution.

Bioinformatics and biomedical informatics with ChatGPT: Year one review

The year 2023 marked a significant surge in the exploration of applying large language model (LLM) chatbots, notably ChatGPT, across various disciplines. We surveyed the applications of ChatGPT in bioinformatics and biomedical informatics throughout the year, covering omics, genetics, biomedical text mining, drug discovery, biomedical image understanding, bioinformatics programming, and bioinformatics education. Our survey delineates the current strengths and limitations of this chatbot in bioinformatics and offers insights into potential avenues for future developments.

nach0: multimodal natural and chemical languages foundation model

Large Language Models (LLMs) have substantially driven scientific progress in various domains, and many papers have demonstrated their ability to tackle complex problems with creative solutions. Our paper introduces a new foundation model, nach0, capable of solving various chemical and biological tasks: biomedical question answering, named entity recognition, molecular generation, molecular synthesis, attributes prediction, and others. nach0 is a multi-domain and multi-task encoder-decoder LLM pre-trained on unlabeled text from scientific literature, patents, and molecule strings to incorporate a range of chemical and linguistic knowledge. We employed instruction tuning, where specific task-related instructions are utilized to fine-tune nach0 for the final set of tasks. To train nach0 effectively, we leverage the NeMo framework, enabling efficient parallel optimization of both base and large model versions. Extensive experiments demonstrate that our model outperforms state-of-the-art baselines on single-domain and cross-domain tasks. Furthermore, it can generate high-quality outputs in molecular and textual formats, showcasing its effectiveness in multi-domain setups.

Performance evaluation of ChatGPT in detecting diagnostic errors and their contributing factors: an analysis of 545 case reports of diagnostic errors

BACKGROUND

Manual chart review using validated assessment tools is a standardised methodology for detecting diagnostic errors. However, this requires considerable human resources and time. ChatGPT, a recently developed artificial intelligence chatbot based on a large language model, can effectively classify text based on suitable prompts. Therefore, ChatGPT can assist manual chart reviews in detecting diagnostic errors.

OBJECTIVE

This study aimed to clarify whether ChatGPT could correctly detect diagnostic errors and possible factors contributing to them based on case presentations.

METHODS

We analysed 545 published case reports that included diagnostic errors. We imputed the texts of case presentations and the final diagnoses with some original prompts into ChatGPT (GPT-4) to generate responses, including the judgement of diagnostic errors and contributing factors of diagnostic errors. Factors contributing to diagnostic errors were coded according to the following three taxonomies: Diagnosis Error Evaluation and Research (DEER), Reliable Diagnosis Challenges (RDC) and Generic Diagnostic Pitfalls (GDP). The responses on the contributing factors from ChatGPT were compared with those from physicians.

RESULTS

ChatGPT correctly detected diagnostic errors in 519/545 cases (95%) and coded statistically larger numbers of factors contributing to diagnostic errors per case than physicians: DEER (median 5 vs 1, p<0.001), RDC (median 4 vs 2, p<0.001) and GDP (median 4 vs 1, p<0.001). The most important contributing factors of diagnostic errors coded by ChatGPT were 'failure/delay in considering the diagnosis' (315, 57.8%) in DEER, 'atypical presentation' (365, 67.0%) in RDC, and 'atypical presentation' (264, 48.4%) in GDP.

CONCLUSION

ChatGPT accurately detects diagnostic errors from case presentations. ChatGPT may be more sensitive than manual reviewing in detecting factors contributing to diagnostic errors, especially for 'atypical presentation'.

Me-LLaMA: Foundation Large Language Models for Medical Applications

Recent advancements in large language models (LLMs) such as ChatGPT and LLaMA have hinted at their potential to revolutionize medical applications, yet their application in clinical settings often reveals limitations due to a lack of specialized training on medical-specific data. In response to this challenge, this study introduces Me-LLaMA, a novel medical LLM family that includes foundation models - Me-LLaMA 13/70B, along with their chat-enhanced versions - Me-LLaMA 13/70B-chat, developed through continual pre-training and instruction tuning of LLaMA2 using large medical datasets. Our methodology leverages a comprehensive domain-specific data suite, including a large-scale, continual pre-training dataset with 129B tokens, an instruction tuning dataset with 214k samples, and a new medical evaluation benchmark (MIBE) across six critical medical tasks with 12 datasets. Our extensive evaluation using the MIBE shows that Me-LLaMA models achieve overall better performance than existing open-source medical LLMs in zero-shot, few-shot and supervised learning abilities. With task-specific instruction tuning, Me-LLaMA models outperform ChatGPT on 7 out of 8 datasets and GPT-4 on 5 out of 8 datasets. In addition, we investigated the catastrophic forgetting problem, and our results show that Me-LLaMA models outperform other open-source medical LLMs in mitigating this issue. Me-LLaMA is one of the largest open-source medical foundation LLMs that use both biomedical and clinical data. It exhibits superior performance across both general and medical tasks compared to other open-source medical LLMs, rendering it an attractive choice for medical AI applications. We release our models, datasets, and evaluation scripts at: https://github.com/BIDS-Xu-Lab/Me-LLaMA.

BiSpec Pairwise AI: guiding the selection of bispecific antibody target combinations with pairwise learning and GPT augmentation

PURPOSE

Bispecific antibodies (BsAbs), capable of targeting two antigens simultaneously, represent a significant advancement by employing dual mechanisms of action for tumor suppression. However, how to pair targets to develop effective and safe bispecific drugs is a major challenge for pharmaceutical companies.

METHODS

Using machine learning models, we refined the biological characteristics of currently approved or in clinical development BsAbs and analyzed hundreds of membrane proteins as bispecific targets to predict the likelihood of successful drug development for various target combinations. Moreover, to enhance the interpretability of prediction results in bispecific target combination, we combined machine learning models with Large Language Models (LLMs). Through a Retrieval-Augmented Generation (RAG) approach, we supplement each pair of bispecific targets' machine learning prediction with important features and rationales, generating interpretable analytical reports.

RESULTS

In this study, the XGBoost model with pairwise learning was employed to predict the druggability of BsAbs. By analyzing extensive data on BsAbs and designing features from perspectives such as target activity, safety, cell type specificity, pathway mechanism, and gene embedding representation, our model is able to predict target combinations of BsAbs with high market potential. Specifically, we integrated XGBoost with the GPT model to discuss the efficacy of each bispecific target pair, thereby aiding the decision-making for drug developers.

CONCLUSION

The novelty of this study lies in the integration of machine learning and GPT techniques to provide a novel framework for the design of BsAbs drugs. This holistic approach not only improves prediction accuracy, but also enhances the interpretability and innovativeness of drug design.

Scientific figures interpreted by ChatGPT: strengths in plot recognition and limits in color perception

Emerging studies underscore the promising capabilities of large language model-based chatbots in conducting basic bioinformatics data analyses. The recent feature of accepting image inputs by ChatGPT, also known as GPT-4V(ision), motivated us to explore its efficacy in deciphering bioinformatics scientific figures. Our evaluation with examples in cancer research, including sequencing data analysis, multimodal network-based drug repositioning, and tumor clonal evolution, revealed that ChatGPT can proficiently explain different plot types and apply biological knowledge to enrich interpretations. However, it struggled to provide accurate interpretations when color perception and quantitative analysis of visual elements were involved. Furthermore, while the chatbot can draft figure legends and summarize findings from the figures, stringent proofreading is imperative to ensure the accuracy and reliability of the content.

Cancer-Alterome: a literature-mined resource for regulatory events caused by genetic alterations in cancer

It is vital to investigate the complex mechanisms underlying tumors to better understand cancer and develop effective treatments. Metabolic abnormalities and clinical phenotypes can serve as essential biomarkers for diagnosing this challenging disease. Additionally, genetic alterations provide profound insights into the fundamental aspects of cancer. This study introduces Cancer-Alterome, a literature-mined dataset that focuses on the regulatory events of an organism's biological processes or clinical phenotypes caused by genetic alterations. By proposing and leveraging a text-mining pipeline, we identify 16,681 thousand of regulatory events records encompassing 21K genes, 157K genetic alterations and 154K downstream bio-concepts, extracted from 4,354K pan-cancer literature. The resulting dataset empowers a multifaceted investigation of cancer pathology, enabling the meticulous tracking of relevant literature support. Its potential applications extend to evidence-based medicine and precision medicine, yielding valuable insights for further advancements in cancer research.

Opportunities and challenges for ChatGPT and large language models in biomedicine and health

ChatGPT has drawn considerable attention from both the general public and domain experts with its remarkable text generation capabilities. This has subsequently led to the emergence of diverse applications in the field of biomedicine and health. In this work, we examine the diverse applications of large language models (LLMs), such as ChatGPT, in biomedicine and health. Specifically, we explore the areas of biomedical information retrieval, question answering, medical text summarization, information extraction and medical education and investigate whether LLMs possess the transformative power to revolutionize these tasks or whether the distinct complexities of biomedical domain presents unique challenges. Following an extensive literature survey, we find that significant advances have been made in the field of text generation tasks, surpassing the previous state-of-the-art methods. For other applications, the advances have been modest. Overall, LLMs have not yet revolutionized biomedicine, but recent rapid progress indicates that such methods hold great potential to provide valuable means for accelerating discovery and improving health. We also find that the use of LLMs, like ChatGPT, in the fields of biomedicine and health entails various risks and challenges, including fabricated information in its generated responses, as well as legal and privacy concerns associated with sensitive patient data. We believe this survey can provide a comprehensive and timely overview to biomedical researchers and healthcare practitioners on the opportunities and challenges associated with using ChatGPT and other LLMs for transforming biomedicine and health.

Bioinformatics Illustrations Decoded by ChatGPT: The Good, The Bad, and The Ugly

Emerging studies underscore the promising capabilities of large language model-based chatbots in conducting fundamental bioinformatics data analyses. The recent feature of accepting image-inputs by ChatGPT motivated us to explore its efficacy in deciphering bioinformatics illustrations. Our evaluation with examples in cancer research, including sequencing data analysis, multimodal network-based drug repositioning, and tumor clonal evolution, revealed that ChatGPT can proficiently explain different plot types and apply biological knowledge to enrich interpretations. However, it struggled to provide accurate interpretations when quantitative analysis of visual elements was involved. Furthermore, while the chatbot can draft figure legends and summarize findings from the figures, stringent proofreading is imperative to ensure the accuracy and reliability of the content.

Data extraction methods for systematic review (semi)automation: Update of a living systematic review

Background

The reliable and usable (semi) automation of data extraction can support the field of systematic review by reducing the workload required to gather information about the conduct and results of the included studies. This living systematic review examines published approaches for data extraction from reports of clinical studies.

Methods

We systematically and continually search PubMed, ACL Anthology, arXiv, OpenAlex via EPPI-Reviewer, and the dblp computer science bibliography databases. Full text screening and data extraction are conducted using a mix of open-source and commercial tools. This living review update includes publications up to August 2024 and OpenAlex content up to September 2024.

Results

117 publications are included in this review. Of these, 30 (26%) used full texts while the rest used titles and abstracts. A total of 112 (96%) publications developed classifiers for randomised controlled trials. Over 30 entities were extracted, with PICOs (population, intervention, comparator, outcome) being the most frequently extracted. Data are available from 53 (45%), and code from 49 (42%) publications. Nine (8%) implemented publicly available tools.

Conclusions

This living systematic review presents an overview of (semi)automated data-extraction literature of interest to different types of literature review. We identified a broad evidence base of publications describing data extraction for interventional reviews and a small number of publications extracting other study types. Between review updates, large language models emerged as a new tool for data extraction. While facilitating access to automated extraction, they showed a trend of decreasing quality of results reporting, especially quantitative results such as recall and lower reproducibility of results. Compared with the previous update, trends such as transition to relation extraction and sharing of code and datasets stayed similar.