preon: Fast and accurate entity normalization for drug names and cancer types in precision oncology

MOTIVATION

In precision oncology (PO), clinicians aim to find the best treatment for any patient based on their molecular characterization. A major bottleneck is the manual annotation and evaluation of individual variants, for which usually a range of knowledge bases are screened. To incorporate and integrate the vast information of different databases, fast and accurate methods for harmonizing databases with different types of information are necessary. An essential step for harmonization in PO includes the normalization of tumor entities as well as therapy options for patients.

SUMMARY

preon is a fast and accurate library for the normalization of drug names and cancer types in large-scale data integration.

AVAILABILITY AND IMPLEMENTATION

preon is implemented in Python and freely available via the PyPI repository. Source code and the data underlying this article are available in GitHub at https://github.com/ermshaua/preon/.

Leveraging Large Language Models for Decision Support in Personalized Oncology

Importance

Clinical interpretation of complex biomarkers for precision oncology currently requires manual investigations of previous studies and databases. Conversational large language models (LLMs) might be beneficial as automated tools for assisting clinical decision-making.

Objective

To assess performance and define their role using 4 recent LLMs as support tools for precision oncology.

Design, Setting, and Participants

This diagnostic study examined 10 fictional cases of patients with advanced cancer with genetic alterations. Each case was submitted to 4 different LLMs (ChatGPT, Galactica, Perplexity, and BioMedLM) and 1 expert physician to identify personalized treatment options in 2023. Treatment options were masked and presented to a molecular tumor board (MTB), whose members rated the likelihood of a treatment option coming from an LLM on a scale from 0 to 10 (0, extremely unlikely; 10, extremely likely) and decided whether the treatment option was clinically useful.

Main Outcomes and Measures

Number of treatment options, precision, recall, F1 score of LLMs compared with human experts, recognizability, and usefulness of recommendations.

Results

For 10 fictional cancer patients (4 with lung cancer, 6 with other; median [IQR] 3.5 [3.0-4.8] molecular alterations per patient), a median (IQR) number of 4.0 (4.0-4.0) compared with 3.0 (3.0-5.0), 7.5 (4.3-9.8), 11.5 (7.8-13.0), and 13.0 (11.3-21.5) treatment options each was identified by the human expert and 4 LLMs, respectively. When considering the expert as a criterion standard, LLM-proposed treatment options reached F1 scores of 0.04, 0.17, 0.14, and 0.19 across all patients combined. Combining treatment options from different LLMs allowed a precision of 0.29 and a recall of 0.29 for an F1 score of 0.29. LLM-generated treatment options were recognized as AI-generated with a median (IQR) 7.5 (5.3-9.0) points in contrast to 2.0 (1.0-3.0) points for manually annotated cases. A crucial reason for identifying AI-generated treatment options was insufficient accompanying evidence. For each patient, at least 1 LLM generated a treatment option that was considered helpful by MTB members. Two unique useful treatment options (including 1 unique treatment strategy) were identified only by LLM.

Conclusions and Relevance

In this diagnostic study, treatment options of LLMs in precision oncology did not reach the quality and credibility of human experts; however, they generated helpful ideas that might have complemented established procedures. Considering technological progress, LLMs could play an increasingly important role in assisting with screening and selecting relevant biomedical literature to support evidence-based, personalized treatment decisions.

Tumour mutational burden and survival with molecularly matched therapy

BACKGROUND

The impact of tumour mutational burden (TMB) on outcome with molecularly matched therapy is unknown. Higher TMB could predict resistance to molecularly matched therapy through co-occurring driver mutations.

METHODS

One hundred and four patients with advanced cancers underwent molecular profiling in the DKTK-MASTER program. Fifty-five patients received systemic therapy excluding immunotherapy. Patients with molecularly matched (n = 35) or non-molecularly informed therapy (n = 20) were analysed for TMB and survival. Results were validated in an independent cohort of patients receiving molecularly matched (n = 68) or non-molecularly informed therapy (n = 40). Co-occurring driver mutations and TMB were analysed in the exploratory cohort and The Cancer Genome Atlas (TCGA) datasets.

RESULTS

Patients were stratified by the median TMB of 1.67 mutations per Megabase (mut/Mb) of 35 patients receiving molecularly matched therapy into TMB-high or TMB-low groups. Median overall survival (4 months [95% CI, 3.3-7.6] versus 12.8 months [95% CI, 10-not reached], p < 0.001) and progression-free survival (1.8 months [95% CI, 1.1-3.7] versus 7.9 months [95% CI, 2.8-17.0], p = 0.003) were significantly shorter in the TMB-high group compared to the TMB-low group. In the validation cohort, shorter OS and PFS were identified in the TMB-high group (TMB cut-off of 4 mut/Mb) treated with molecularly matched therapy. No differences were observed in patients receiving non-molecularly informed systemic therapy. A significant correlation between co-occurring driver mutations and TMB (n = 104, r = 0.78 [95% CI, 0.68-0.85], p < 0.001) was found in the exploratory cohort as well as the majority (24/33) of TCGA studies.

CONCLUSION

A high TMB was associated with unfavourable outcome in patients receiving molecularly matched therapy, indicating untargeted resistance pathways. Therefore, TMB should be further investigated as a predictive biomarker in precision oncology programs.

Creation of a structured molecular genomics report for Germany as a local adaption of HL7's Genomic Reporting Implementation Guide

OBJECTIVE

The objective was to develop a dataset definition, information model, and FHIR® specification for key data elements contained in a German molecular genomics (MolGen) report to facilitate genomic and phenotype integration in electronic health records.

MATERIALS AND METHODS

A dedicated expert group participating in the German Medical Informatics Initiative reviewed information contained in MolGen reports, determined the key elements, and formulated a dataset definition. HL7's Genomics Reporting Implementation Guide (IG) was adopted as a basis for the FHIR® specification which was subjected to a public ballot. In addition, elements in the MolGen dataset were mapped to the fields defined in ISO/TS 20428:2017 standard to evaluate compliance.

RESULTS

A core dataset of 76 data elements, clustered into 6 categories was created to represent all key information of German MolGen reports. Based on this, a FHIR specification with 16 profiles, 14 derived from HL7®'s Genomics Reporting IG and 2 additional profiles (of the FamilyMemberHistory and RiskAssessment resources), was developed. Five example resource bundles show how our adaptation of an international standard can be used to model MolGen report data that was requested following oncological or rare disease indications. Furthermore, the map of the MolGen report data elements to the fields defined by the ISO/TC 20428:2017 standard, confirmed the presence of the majority of required fields.

CONCLUSIONS

Our report serves as a template for other research initiatives attempting to create a standard format for unstructured genomic report data. Use of standard formats facilitates integration of genomic data into electronic health records for clinical decision support.

Feasibility and outcome of reproducible clinical interpretation of high-dimensional molecular data: a comparison of two molecular tumor boards

BACKGROUND

Structured and harmonized implementation of molecular tumor boards (MTB) for the clinical interpretation of molecular data presents a current challenge for precision oncology. Heterogeneity in the interpretation of molecular data was shown for patients even with a limited number of molecular alterations. Integration of high-dimensional molecular data, including RNA- (RNA-Seq) and whole-exome sequencing (WES), is expected to further complicate clinical application. To analyze challenges for MTB harmonization based on complex molecular datasets, we retrospectively compared clinical interpretation of WES and RNA-Seq data by two independent molecular tumor boards.

METHODS

High-dimensional molecular cancer profiling including WES and RNA-Seq was performed for patients with advanced solid tumors, no available standard therapy, ECOG performance status of 0-1, and available fresh-frozen tissue within the DKTK-MASTER Program from 2016 to 2018. Identical molecular profiling data of 40 patients were independently discussed by two molecular tumor boards (MTB) after prior annotation by specialized physicians, following independent, but similar workflows. Identified biomarkers and resulting treatment options were compared between the MTBs and patients were followed up clinically.

RESULTS

A median of 309 molecular aberrations from WES and RNA-Seq (n = 38) and 82 molecular aberrations from WES only (n = 3) were considered for clinical interpretation for 40 patients (one patient sequenced twice). A median of 3 and 2 targeted treatment options were identified per patient, respectively. Most treatment options were identified for receptor tyrosine kinase, PARP, and mTOR inhibitors, as well as immunotherapy. The mean overlap coefficient between both MTB was 66%. Highest agreement rates were observed with the interpretation of single nucleotide variants, clinical evidence levels 1 and 2, and monotherapy whereas the interpretation of gene expression changes, preclinical evidence levels 3 and 4, and combination therapy yielded lower agreement rates. Patients receiving treatment following concordant MTB recommendations had significantly longer overall survival than patients receiving treatment following discrepant recommendations or physician's choice.

CONCLUSIONS

Reproducible clinical interpretation of high-dimensional molecular data is feasible and agreement rates are encouraging, when compared to previous reports. The interpretation of molecular aberrations beyond single nucleotide variants and preclinically validated biomarkers as well as combination therapies were identified as additional difficulties for ongoing harmonization efforts.

Standards for the classification of pathogenicity of somatic variants in cancer (oncogenicity): Joint recommendations of Clinical Genome Resource (ClinGen), Cancer Genomics Consortium (CGC), and Variant Interpretation for Cancer Consortium (VICC)

PURPOSE

Several professional societies have published guidelines for the clinical interpretation of somatic variants, which specifically address diagnostic, prognostic, and therapeutic implications. Although these guidelines for the clinical interpretation of variants include data types that may be used to determine the oncogenicity of a variant (eg, population frequency, functional, and in silico data or somatic frequency), they do not provide a direct, systematic, and comprehensive set of standards and rules to classify the oncogenicity of a somatic variant. This insufficient guidance leads to inconsistent classification of rare somatic variants in cancer, generates variability in their clinical interpretation, and, importantly, affects patient care. Therefore, it is essential to address this unmet need.

METHODS

Clinical Genome Resource (ClinGen) Somatic Cancer Clinical Domain Working Group and ClinGen Germline/Somatic Variant Subcommittee, the Cancer Genomics Consortium, and the Variant Interpretation for Cancer Consortium used a consensus approach to develop a standard operating procedure (SOP) for the classification of oncogenicity of somatic variants.

RESULTS

This comprehensive SOP has been developed to improve consistency in somatic variant classification and has been validated on 94 somatic variants in 10 common cancer-related genes.

CONCLUSION

The comprehensive SOP is now available for classification of oncogenicity of somatic variants.

Disentangling Pro-mitotic Signaling during Cell Cycle Progression using Time-Resolved Single-Cell Imaging

Cells rely on input from extracellular growth factors to control their proliferation during development and adult homeostasis. Such mitogenic inputs are transmitted through multiple signaling pathways that synergize to precisely regulate cell cycle entry and progression. Although the architecture of these signaling networks has been characterized in molecular detail, their relative contribution, especially at later cell cycle stages, remains largely unexplored. By combining quantitative time-resolved measurements of fluorescent reporters in untransformed human cells with targeted pharmacological inhibitors and statistical analysis, we quantify epidermal growth factor (EGF)-induced signal processing in individual cells over time and dissect the dynamic contribution of downstream pathways. We define signaling features that encode information about extracellular ligand concentrations and critical time windows for inducing cell cycle transitions. We show that both extracellular signal-regulated kinase (ERK) and phosphatidylinositol 3-kinase (PI3K) activity are necessary for initial cell cycle entry, whereas only PI3K affects the duration of S phase at later stages of mitogenic signaling.

Support of a molecular tumour board by an evidence-based decision management system for precision oncology

BACKGROUND

Reliable and reproducible interpretation of molecular aberrations constitutes a bottleneck of precision medicine. Evidence-based decision management systems may improve rational therapy recommendations. To cope with an increasing amount of complex molecular data in the clinical care of patients with cancer, we established a workflow for the interpretation of molecular analyses.

METHODS

A specialized physician screened results from molecular analyses for potential biomarkers, irrespective of the diagnostic modality. Best available evidence was retrieved and categorized through establishment of an in-house database and interrogation of publicly available databases. Annotated biomarkers were ranked using predefined evidence levels and subsequently discussed at a molecular tumour board (MTB), which generated treatment recommendations. Subsequent translation into patient treatment and clinical outcomes were followed up.

RESULTS

One hundred patients were discussed in the MTB between January 2016 and May 2017. Molecular data were obtained for 70 of 100 patients (50 whole exome/RNA sequencing, 18 panel sequencing, 2 immunohistochemistry (IHC)/microsatellite instability analysis). The MTB generated a median of two treatment recommendations each for 63 patients. Thirty-nine patients were treated: 6 partial responses and 12 stable diseases were achieved as best responses. Genetic counselling for germline events was recommended for seven patients.

CONCLUSION

The development of an evidence-based workflow allowed for the clinical interpretation of complex molecular data and facilitated the translation of personalized treatment strategies into routine clinical care. The high number of treatment recommendations in patients with comprehensive genomic data and promising responses in patients treated with combination therapy warrant larger clinical studies.

Comparative Analysis of Public Knowledge Bases for Precision Oncology

PURPOSE

Precision oncology depends on the availability of up-to-date, comprehensive, and accurate information about associations between genetic variants and therapeutic options. Recently, a number of knowledge bases (KBs) have been developed that gather such information on the basis of expert curation of the scientific literature. We performed a quantitative and qualitative comparison of Clinical Interpretations of Variants in Cancer, OncoKB, Cancer Gene Census, Database of Curated Mutations, CGI Biomarkers (the cancer genome interpreter biomarker database), Tumor Alterations Relevant for Genomics-Driven Therapy, and the Precision Medicine Knowledge Base.

METHODS

We downloaded each KB and restructured their content to describe variants, genes, drugs, and gene-drug associations in a common format. We normalized gene names to Entrez Gene IDs and drug names to ChEMBL and DrugBank IDs. For the analysis of clinically relevant gene-drug associations, we obtained lists of genes affected by genetic alterations and putative drug therapies for 113 patients with cancer whose cases were presented at the Molecular Tumor Board (MTB) of the Charité Comprehensive Cancer Center.

RESULTS

Our analysis revealed that the KBs are largely overlapping but also that each source harbors a notable amount of unique information. Although some KBs cover more genes, others contain more data about gene-drug associations. Retrospective comparisons with findings of the Charitè MTB at the gene level showed that use of multiple KBs may considerably improve retrieval results. The relative importance of a KB in terms of cancer genes was assessed in more detail by logistic regression, which revealed that all but one source had a notable impact on result quality. We confirmed these findings using a second data set obtained from an independent MTB.

CONCLUSION

To date, none of the existing publicly available KBs on gene-drug associations in precision oncology fully subsumes the others, but all of them exhibit specific strengths and weaknesses. Consideration of multiple KBs, therefore, is essential to obtain comprehensive results.

Clinical impact of comprehensive versus targeted genomic analysis for precision oncology

e13033

Background: Panel sequencing (PS) has become a standard-of-care in cancer diagnostics. More comprehensive analyses such as whole-exome (WES) or RNA sequencing (RNAseq) allow for the detection of rare and unknown genetic aberrations that are not covered by predefined assays. The clinical impact of targeted versus comprehensive genomic assays were analyzed in patients presented at the Charité Molecular Tumor Board (MTB). Methods: Patients (pts) with advanced and/or metastatic cancer for whom no standard therapy was available were discussed in the MTB to allocate diagnostic profiling and guide biomarker-based treatment (BBT). Pts had to be < 50 years of age or diagnosed with a rare tumor entity to undergo WES/RNAseq, performed on fresh tissue. If ineligible, standard PS was performed on archival tissue. BBT recommendations, ranked by pre-specified evidence levels, were made by the MTB and pts were followed up. Results: 228 patients (median age 49 years, 108 female and 120 male) were discussed in the MTB between January 2016 and February 2019. We assigned 73 and 155 pts to PS and WES/RNAseq and results were obtained for 78.1% (n = 57/73) and 54.8% (n = 85/155) pts, respectively. Sequencing failed for 11 (PS; 15.1%) and 62 (WES/RNAseq; 40%) pts, most commonly due to insufficient tissue (n = 29). Sequencing was ongoing in 5 (PS) and 8 (WES/RNAseq) pts at the time of analysis. A median of 2 BBTs were recommended for 75.4% (43/57) of PS (range r: 1-3) and 90.6% (77/85) of WES/RNAseq pts (r: 1-6) each. 22% (n = 17/77) of WES/RNAseq pts had ≥4 BBTs made by the MTB. Treatment was initiated in 30.2% (n = 13/43) of PS and 40.2% (n = 31/77) of WES/RNAseq pts. Clinical benefit rates (CBRs) were 23.1% (2 PR, 1 SD) for PS and 45.2% (2 CR, 3 PR, 9 SD) for WES/RNAseq pts. Overall survival data was immature at the time of analysis. Conclusions: Utilizing WES/RNAseq is a feasible approach to perform tumor profiling in a heterogeneous cohort. We here show a higher rate of pts receiving confident evidence-based treatment recommendations in the WES/RNAseq group and a higher rate of treatment initiation. The CBR nearly doubled in the WES/RNAseq cohort when compared to standard PS pts, thus emphasizing the need for larger comparative analyses to guide diagnostic decision-making.

How cyanobacteria pose new problems to old methods: challenges in microarray time series analysis

Background

The transcriptomes of several cyanobacterial strains have been shown to exhibit diurnal oscillation patterns, reflecting the diurnal phototrophic lifestyle of the organisms. The analysis of such genome-wide transcriptional oscillations is often facilitated by the use of clustering algorithms in conjunction with a number of pre-processing steps. Biological interpretation is usually focussed on the time and phase of expression of the resulting groups of genes. However, the use of microarray technology in such studies requires the normalization of pre-processing data, with unclear impact on the qualitative and quantitative features of the derived information on the number of oscillating transcripts and their respective phases.

Results

A microarray based evaluation of diurnal expression in the cyanobacterium Synechocystis sp. PCC 6803 is presented. As expected, the temporal expression patterns reveal strong oscillations in transcript abundance. We compare the Fourier transformation-based expression phase before and after the application of quantile normalization, median polishing, cyclical LOESS, and least oscillating set (LOS) normalization. Whereas LOS normalization mostly preserves the phases of the raw data, the remaining methods introduce systematic biases. In particular, quantile-normalization is found to introduce a phase-shift of 180°, effectively changing night-expressed genes into day-expressed ones. Comparison of a large number of clustering results of differently normalized data shows that the normalization method determines the result. Subsequent steps, such as the choice of data transformation, similarity measure, and clustering algorithm, only play minor roles. We find that the standardization and the DTF transformation are favorable for the clustering of time series in contrast to the 12 m transformation. We use the cluster-wise functional enrichment of a clustering derived by LOS normalization, clustering using flowClust, and DFT transformation to derive the diurnal biological program of Synechocystis sp..

Conclusion

Application of quantile normalization, median polishing, and also cyclic LOESS normalization of the presented cyanobacterial dataset lead to increased numbers of oscillating genes and the systematic shift of the expression phase. The LOS normalization minimizes the observed detrimental effects. As previous analyses employed a variety of different normalization methods, a direct comparison of results must be treated with caution.

IL-2 Expression in Activated Human Memory FOXP3(+) Cells Critically Depends on the Cellular Levels of FOXP3 as Well as of Four Transcription Factors of T Cell Activation

The human CD4⁺FOXP3⁺ T cell population is heterogeneous and consists of various subpopulations which remain poorly defined. Anergy and suppression are two main functional characteristics of FOXP3⁺Treg cells. We used the anergic behavior of FOXP3⁺Treg cells for a better discrimination and characterization of such subpopulations. We compared IL-2-expressing with IL-2-non-expressing cells within the memory FOXP3⁺ T cell population. In contrast to IL-2-non-expressing FOXP3⁺ cells, IL-2-expressing FOXP3⁺ cells exhibit intermediate characteristics of Treg and Th cells concerning the Treg cell markers CD25, GITR, and Helios. Besides lower levels of FOXP3, they also have higher levels of the transcription factors NFATc2, c-Fos, NF-κBp65, and c-Jun. An approach combining flow cytometric measurements with statistical interpretation for quantitative transcription factor analysis suggests that the physiological expression levels not only of FOXP3 but also of NFATc2, c-Jun, c-Fos, and NF-κBp65 are limiting for the decision whether IL-2 is expressed or not in activated peripheral human memory FOXP3⁺ cells. These findings demonstrate that concomitant high levels of NFATc2, c-Jun, c-Fos, and NF-κBp65 lead in addition to potential IL-2 expression in those FOXP3⁺ cells with low levels of FOXP3. We hypothesize that not only the level of FOXP3 expression but also the amounts of the four transcription factors studied represent determining factors for the anergic phenotype of FOXP3⁺ Treg cells.

Stable IL-2 decision making by endogenous c-Fos amounts in peripheral memory T-helper cells

The cytokine IL-2 performs opposite functions supporting efficient immune responses and playing a key role in peripheral tolerance. Therefore, precise fine-tuning of IL-2 expression is crucial for adjusting the immune response. Combining transcription factor analysis at the single cell and the single nucleus level using flow cytometry with statistical analysis, we showed that physiological differences in the expression levels of c-Fos and NFATc2, but not of c-Jun and NF-κBp65, are limiting for the decision whether IL-2 is expressed in a strongly activated human memory T-helper (Th) cell. Variation in the expression of c-Fos leads to substantial diversity of IL-2 expression in ∼40% of the memory Th cells. The remaining cells exhibit an equally high c-Fos expression level, thereby ensuring robustness in IL-2 response within the population. These findings reveal how memory Th cells benefit from regulated variation in transcription factor expression to achieve a certain stability and variability of cytokine expression in a controlled manner.

Prediction of regulatory transcription factors in T helper cell differentiation and maintenance

Naive T-helper (Th) cells differentiate into distinct lineages including Th1, Th2, Th17 and regulatory T (Treg) cells. Each of these Th-lineages has specific functions in immune defense and T cell homeostasis. Th cell fate decisions and commitment are dependent on the kind and strength of T cell stimulation and the subsequent gene expression profiles. Our analysis targeted the identification of new regulatory transcription factor binding sites (TFBSs) in the promoter regions of up- and down-regulated genes in Treg cell differentiation and lineage maintenance. For this approach we compared different gene groups from global gene expression studies with background models of randomly selected genes to identify significantly overrepresented TFBSs. Results of our analysis suggest that Ets and IRF family members contribute to the regulation of the initial induction of Treg cells. Furthermore, we identified the overrepresented TFBS-pairs Runx-NFAT and GATA3-Foxp3 in Treg specific genes and Foxp3 dependent genes, respectively. Interestingly, previous studies have observed functional interactions of both TFBS-pairs in T cells. This study provides a starting point for further investigations to elucidate the transcriptional network in Treg cells.

Modeling IL-2 gene expression in human regulatory T cells

Interleukin-2 (IL-2) is one of the first cytokines to be expressed by T helper cells (Th cells) after antigen-specific stimulation. In contrast, regulatory T cells (T(reg) cells) do not express IL-2, although they are activated via the same pathways. In regulatory T cells the additional transcription factor FoxP3 is expressed. Using intracellular measurement of the transcription factors NFAT and FoxP3 as well as the cytokine IL-2 on single cell level we revealed a small fraction of IL-2 expressing T(reg) cells. Furthermore, these data enabled us to develop initial mathematical models describing gene expression of IL-2 in individual cells. The models are adapted to data from human regulatory T cells. Based on statistical tests of available flow cytometric data it seems reasonable that not only the amount of the transcription factors NFAT and FoxP3 is important but also their concentration ratio. We discuss specific problems of modeling gene expression on single cell level taking IL-2 expression as an example.