A comprehensive evaluation of regression-based drug responsiveness prediction models, using cell viability inhibitory concentrations (IC50 values)

Constructing a prognostic model for osteosarcoma based on centrosome-related genes and identifying potential therapeutic targets of paclitaxel

The centrosome, a vital component in mitosis in eukaryotes, plays a pivotal role in cancer progression by influencing the proliferation and differentiation of malignant cells, making it a significant therapeutic target. We collected genes associated with centrosomes from existing literature and established a prognostic model for 85 osteosarcoma patients from the TARGET database. Genes associated with prognosis were identified through univariate Cox regression. We then mitigated overfitting by addressing collinearity using LASSO regression. Ultimately, a set of five genes was selected for the model through multivariable Cox regression. Model performance was assessed using ROC curves, which yielded a training set AUC of 0.965 and a validation set AUC of 0.770, indicating satisfactory model performance. We further identified genes with differential expression in high and low-risk groups and conducted functional enrichment analysis using KEGG, GO, Progeny, GSVA, and GSEA. Results revealed significant variances in various immune-related pathways between high and low-risk cohorts. Analysis of the immune microenvironment using ssGSEA and ESTIMATE indicated that individuals with unfavorable prognoses had lower immune scores, stromal scores, and ESTIMATE scores, coupled with higher tumor purity. This suggests that high-risk individuals have compromised immune microenvironments, potentially contributing to their unfavorable prognoses. Additionally, drug sensitivity and molecular docking analysis revealed increased responsiveness to paclitaxel in high-risk individuals, implying its prognostic value. The JTB-encoded protein exhibited a negative binding energy of - 5.5 kcal/mol when interacting with paclitaxel, indicating its potential to enhance the patient's immune microenvironment. This framework enables patient prognosis prediction and sheds light on paclitaxel's mechanism in osteosarcoma treatment, facilitating personalized treatment approaches.

Development of a breast cancer invasion score to predict tumor aggressiveness and prognosis via PI3K/AKT/mTOR pathway analysis

Invasiveness is a key indicator of tumor malignancy and is often linked to poor prognosis in breast cancer (BC). To explore the diverse characteristics of invasive cells, single-cell RNA sequencing (scRNA-seq) data from three ductal carcinoma stages were analyzed, classifying samples into invasion and non-invasion groups. Nine genes (MCTS1, PGK1, PCMT1, C8orf76, TMEM242, QPRT, SLC16A2, AFG1L, and SPINK8) were identified as key discriminators between these groups. A breast cancer invasion score (BCIS) model was developed using LASSO Cox regression, revealed that high BCIS correlated with poorer overall survival in TCGA-BRCA patients and was validated across GSE20685 and METABRIC datasets (five-year and ten-year survival). Functional experiments demonstrated that knockdown of PGK1 or PCMT1 inhibited tumor cell proliferation and reduced the phosphorylation levels of mTORC, P70S6K, S6, and AKT, indicating suppression of the PI3K/AKT/mTOR pathways. High-BCIS tumors exhibited enrichment in protein secretion and PI3K/AKT/mTOR pathways, associated with aggressiveness and therapy resistance. This study introduced the BCIS score, distinguishing invasion from non-invasion cells, linked to PI3K/AKT/mTOR pathways, offering insights into BRCA prognosis and tumor aggressiveness.

Height estimation in children and adolescents using body composition big data: Machine-learning and explainable artificial intelligence approach

Objective

To develop an accurate and interpretable height estimation model for children and adolescents using body composition variables and explainable artificial intelligence approaches.

Methods

A light gradient boosting method was employed on a dataset of 278,301 measurements from 54,374 children and adolescents aged 6-18 years. The model incorporated anthropometric and body composition measures. Model interpretability was enhanced through feature importance analysis, Shapley additive explanations, partial dependence plots, and accumulated local effects.

Results

The models achieved high accuracy with mean absolute percentage errors of 1.64% and 1.63% for boys and girls, respectively. Soft lean mass (SLM), body fat mass percentage (BFMP), skeletal muscle mass, and skeletal muscle mass percentage were consistently identified as key factors influencing height estimation. Analysis revealed a positive correlation between SLM and estimated height, while BFMP exhibited an inverse relationship with height projections.

Conclusion

These findings provide valuable insights into the relationship between body composition and height, underlining the potential of body composition variables as accurate height predictors in children and adolescents. The model's interpretability and accuracy make it a promising tool for pediatric growth assessment and monitoring.

Synthesis and antitumor activity of copper(II) complexes of imidazole derivatives

Complexes [Cu(PI)2(H2O)](NO3)2 (1), [Cu(PBI)2(NO3)]NO3 (2), [Cu(TBI)2(NO3)]NO3 (3), [Cu(BBIP)2](ClO4)2 (4) and [Cu(BBIP)(CH3OH)(ClO4)2] (5) were synthesized from the reactions of Cu(II) salts with 2-(2'-pyridyl)imidazole (PI), (2-(2'-pyridyl)benzimidazole (PBI), 2-(4'-thiazolyl)-benzimidazole (TBI), 2,6-bis(benzimidazol-2-yl)-pyridine (BBIP), respectively. Their compositions and crystal structures were determined. Their in-vitro antitumor activities were screened on four cancer cell lines and one normal cell line (HL-7702) using cisplatin as the positive control. Complexes 2 and 4 show higher cytotoxicity than the other three complexes. The cytotoxicity of complex 2 are comparable to those for cisplatin, and the cytotoxicity for 4 are much higher than those for cisplatin. From a viewpoint of antitumor, 2 might be a nice choice on the tumor cell line of T24 because its IC50 values on T24 and HL-7702 are 15.03 ± 1.10 and 21.34 ± 0.35, respectively. Thus, a mechanistic study for complexes 2 and 4 on T24 cells was conducted. It revealed that they can reduce mitochondrial membrane potential and increase mitochondrial membrane permeability, resulting in increased intracellular ROS levels, Ca2+ inward flow, dysfunctional mitochondria and the eventual cell apoptosis. In conclusion, they can induce cell apoptosis through mitochondrial dysfunction. These findings could be useful in the development of new antitumor agents.

Decoding the impact of neighboring amino acids on ESI-MS intensity output through deep learning

Peptide-level quantification using mass spectrometry (MS) is no trivial task as the physicochemical properties affect both response and detectability. The specific amino acid (AA) sequence affects these properties, however the connection between sequence and intensity output remains poorly understood. In this work, we explore combinations of amino acid pairs (i.e., dimer motifs) to determine a potential relationship between the local amino acid environment and MS1 intensity. For this purpose, a deep learning (DL) model, consisting of an encoder-decoder with an attention mechanism, was built. The attention mechanism allowed to identify the most relevant motifs. Specific patterns were consistently observed where a bulky/aromatic and hydrophobic AA followed by a cationic AA as well as consecutive bulky/aromatic and hydrophobic AAs were found important for the prediction of the MS1 intensity. Correlating attention weights to mean MS1 intensities revealed that some important motifs, particularly containing Trp, His, and Cys, were linked with low responding peptides whereas motifs containing Lys and most bulky hydrophobic AAs were often associated with high responding peptides. Moreover, Asn-Gly was associated with low response. The model predicts MS1 response with a mean average percentage error of ∼11 % and a Pearson correlation coefficient of ∼0.64. While dimer representation of peptide sequences did not improve predictive capacity compared to single AA representation in earlier work, this work adds valuable insight for a better understanding of peptide response in MS analysis. SIGNIFICANCE: Mass spectrometry is not inherently quantitative, and the response of a compound relies not only on its concentration but also on the molecular composition. For mass spectrometry-based analysis of peptides, such as in bottom-up proteomics, this directly implies that the response cannot be used directly to quantify individual peptides. Moreover, the dependency of the response on the amino acid sequence of individual peptides remains poorly understood. Using a deep learning model based on a recurrent neural network with an attention mechanism, we here investigate how the presence of dimer motifs within a peptide affects the MS1 response through the analysis of intended equimolar peptide pools comprising almost 200,000 unique peptides in total. Not only do we identify certain dimer classes and specific dimers that substantially affect the MS1 response, but the model is also able to predict peptide intensity with low error rates within the independent test subset. The findings not only improve our understanding of the link between sequence and response for peptides but also highlight the potential of utilizing deep learning for developing methods allowing for absolute, label-free peptide quantification.

Machine learning in oncological pharmacogenomics: advancing personalized chemotherapy

This review analyzes the application of machine learning (ML) in oncological pharmacogenomics, focusing on customizing chemotherapy treatments. It explores how ML can analyze extensive genomic, proteomic, and other omics datasets to identify genetic patterns associated with drug responses. This, in turn, facilitates personalized therapies that are more effective and have fewer side effects. Recent studies have emphasized ML's revolutionary role of ML in personalized oncology treatment by identifying genetic variability and understanding cancer pharmacodynamics. Integrating ML with electronic health records and clinical data shows promise in refining chemotherapy recommendations by considering the complex influencing factors. Although standard chemotherapy depends on population-based doses and treatment regimens, customized techniques use genetic information to tailor treatments for specific patients, potentially enhancing efficacy and reducing adverse effects.However, challenges, such as model interpretability, data quality, transparency, ethical issues related to data privacy, and health disparities, remain. Machine learning has been used to transform oncological pharmacogenomics by enabling personalized chemotherapy treatments. This review highlights ML's potential of ML to enhance treatment effectiveness and minimize side effects through detailed genetic analysis. It also addresses ongoing challenges including improved model interpretability, data quality, and ethical considerations. The review concludes by emphasizing the importance of rigorous clinical trials and interdisciplinary collaboration in the ethical implementation of ML-driven personalized medicine, paving the way for improved outcomes in cancer patients and marking a new frontier in cancer treatment.

PKMYT1 knockdown inhibits cholesterol biosynthesis and promotes the drug sensitivity of triple-negative breast cancer cells to atorvastatin

Triple negative breast cancer (TNBC) as the most aggressive molecular subtype of breast cancer is characterized by high cancer cell proliferation and poor patient prognosis. Abnormal lipid metabolism contributes to the malignant process of cancers. Study observed significantly enhanced cholesterol biosynthesis in TNBC. However, the mechanisms underlying the abnormal increase of cholesterol biosynthesis in TNBC are still unclear. Hence, we identified a member of the serine/threonine protein kinase family PKMYT1 as a key driver of cholesterol synthesis in TNBC cells. Aberrantly high-expressed PKMYT1 in TNBC was indicative of unfavorable prognostic outcomes. In addition, PKMYT1 promoted sterol regulatory element-binding protein 2 (SREBP2)-mediated expression of enzymes related to cholesterol biosynthesis through activating the TNF/ TNF receptor-associated factor 1 (TRAF1)/AKT pathway. Notably, downregulation of PKMYT1 significantly inhibited the feedback upregulation of statin-mediated cholesterol biosynthesis, whereas knockdown of PKMYT1 promoted the drug sensitivity of atorvastatin in TNBC cells. Overall, our study revealed a novel function of PKMYT1 in TNBC cholesterol biosynthesis, providing a new target for targeting tumor metabolic reprogramming in the cancer.

Deconvoluting low yield from weak potency in direct-to-biology workflows with machine learning

High throughput and rapid biological evaluation of small molecules is an essential factor in drug discovery and development. Direct-to-biology (D2B), whereby compound purification is foregone, has emerged as a viable technique in time efficient screening, specifically for PROTAC design and biological evaluation. However, one notable limitation is the prerequisite of high yielding reactions to ensure the desired compound is indeed the compound responsible for biological activity. Herein, we report a machine learning based yield-assay deconfounder capable of deconvoluting low yield from low potency to identify false negatives. We validated this approach by identifying promising SARS-CoV-2 main protease inhibitors with nanomolar activity that rivaled potency observed from the standard D2B workflow. Furthermore, we show how our framework can be utilized in a broad, in silico screen to produce compounds of similar potency as a D2B assay.

Matrix stiffness triggers chemoresistance through elevated autophagy in pancreatic ductal adenocarcinoma

Pancreatic ductal adenocarcinoma (PDAC) has a signature of extremely high matrix stiffness caused by a special desmoplastic reaction, which dynamically stiffens along with the pathological process. The poor prognosis and low five-year survival rate of PDAC are partly owing to chemoresistance triggered by substrate stiffness. Understanding the potential mechanisms of matrix stiffness causing PDAC chemoresistance is of great significance. In this study, methacrylated gelatin hydrogel was used as platform for PANC-1 and MIA-PaCa2 cell culture. The results indicated that compared to soft substrate, stiff substrate distinctively reduced the gemcitabine sensitivity of pancreatic cancer. Intriguingly, transmission electron microscopy, immunofluorescence staining, western blot and qRT-PCR assay showcased that the number of autophagosomes and the expression of LC3 were elevated. The observations indicate that matrix stiffness may regulate the autophagy level, which plays a vital role during chemoresistance. In brief, soft substrate exhibited low autophagy level, while the counterpart displayed elevated autophagy level. In order to elucidate the underlying interaction between matrix stiffness-mediated cell autophagy and chemoresistance, rescue experiments with rapamycin and chloroquine were conducted. We found that inhibiting cell autophagy dramatically increased the sensitivity of pancreatic cancer cells to gemcitabine in the stiff group, while promoting autophagy-driven chemoresistance in the soft group, demonstrating that matrix stiffness modulated chemoresistance via autophagy. Furthermore, RNA-seq results showed that miR-1972 may regulate autophagy level in response to matrix stiffness. Overall, our research shed light on the synergistic therapy of PDAC combined with gemcitabine and chloroquine, which is conducive to promoting a therapeutic effect.

A performance evaluation of drug response prediction models for individual drugs

Drug response prediction is important to establish personalized medicine for cancer therapy. Model construction for predicting drug response (i.e., cell viability half-maximal inhibitory concentration [IC₅₀]) of an individual drug by inputting pharmacogenomics in disease models remains critical. Machine learning (ML) has been predominantly applied for prediction, despite the advent of deep learning (DL). Moreover, whether DL or traditional ML models are superior for predicting cell viability IC_50s has to be established. Herein, we constructed ML and DL drug response prediction models for 24 individual drugs and compared the performance of the models by employing gene expression and mutation profiles of cancer cell lines as input. We observed no significant difference in drug response prediction performance between DL and ML models for 24 drugs [root mean squared error (RMSE) ranging from 0.284 to 3.563 for DL and from 0.274 to 2.697 for ML; R² ranging from -7.405 to 0.331 for DL and from -8.113 to 0.470 for ML]. Among the 24 individual drugs, the ridge model of panobinostat exhibited the best performance (R² 0.470 and RMSE 0.623). Thus, we selected the ridge model of panobinostat for further application of explainable artificial intelligence (XAI). Using XAI, we further identified important genomic features for panobinostat response prediction in the ridge model, suggesting the genomic features of 22 genes. Based on our findings, results for an individual drug employing both DL and ML models were comparable. Our study confirms the applicability of drug response prediction models for individual drugs.

Insight on physicochemical properties governing peptide MS1 response in HPLC-ESI-MS/MS: A deep learning approach

Accurate and absolute quantification of peptides in complex mixtures using quantitative mass spectrometry (MS)-based methods requires foreground knowledge and isotopically labeled standards, thereby increasing analytical expenses, time consumption, and labor, thus limiting the number of peptides that can be accurately quantified. This originates from differential ionization efficiency between peptides and thus, understanding the physicochemical properties that influence the ionization and response in MS analysis is essential for developing less restrictive label-free quantitative methods. Here, we used equimolar peptide pool repository data to develop a deep learning model capable of identifying amino acids influencing the MS1 response. By using an encoder-decoder with an attention mechanism and correlating attention weights with amino acid physicochemical properties, we obtain insight on properties governing the peptide-level MS1 response within the datasets. While the problem cannot be described by one single set of amino acids and properties, distinct patterns were reproducibly obtained. Properties are grouped in three main categories related to peptide hydrophobicity, charge, and structural propensities. Moreover, our model can predict MS1 intensity output under defined conditions based solely on peptide sequence input. Using a refined training dataset, the model predicted log-transformed peptide MS1 intensities with an average error of 9.7 ± 0.5% based on 5-fold cross validation, and outperformed random forest and ridge regression models on both log-transformed and real scale data. This work demonstrates how deep learning can facilitate identification of physicochemical properties influencing peptide MS1 responses, but also illustrates how sequence-based response prediction and label-free peptide-level quantification may impact future workflows within quantitative proteomics.

GlucoKinaseDB: A comprehensive, curated resource of glucokinase modulators for clinical and molecular research

Glucokinase (GK), an isoform of hexokinase expressed predominantly in liver, pancreas and hypothalamus is crucial to blood glucose management. It is a critical component of the glucose-sensing mechanism of the pancreatic islet cells and glycogen regulation in hepatocytes. GK modulators such as allosteric GKAs (glucokinase activators) and GK-GKRP (glucokinase regulatory protein) disruptors have found potential applications as safer antihyperglycemics. Recent studies have also demonstrated the potential of GK modulators as antiparasitic agents. Researchers targeting GK often undertake the time-consuming task of independently collecting and compiling modulator information due to the lack of any dedicated single-platform resource. Towards this, in the present study we demonstrate the design and development of GlucoKinaseDB (GKDB), a comprehensive, curated, online resource of GK modulators. GKDB contains experimentally derived structural and bioactivity information of 1723 modulators along with their detailed molecular descriptors. The web-interface is user-friendly with features such as in-browser visualization, advanced search queries, cross-links to other databases and original reference etc. The bioactivity and descriptor data can be downloaded in bulk (for entire database) or for individual modulators. The 3D structures are also downloadable in multiple formats. GKDB employs a PHP-based web design with Bootstrap styling and a MySQL database backend. GKDB can be utilized for clinical and molecular research via development of pharmacophore hypotheses, QSAR/QSPR models, predictive machine learning models etc. GKDB is freely accessible online at https://glucokinasedb.in.

Oncogenic signaling pathways and hallmarks of cancer in Korean patients with acral melanoma

Acral melanoma (AM), a rare subtype of cutaneous melanoma, shows higher incidence in Asians, including Koreans, than in Caucasians. However, the genetic modification associated with AM in Koreans is not well known and has not been comprehensively investigated in terms of oncogenic signaling, and hallmarks of cancer. We performed whole-exome and RNA sequencing for Korean patients with AM and acquired the genetic alterations and gene expression profiles. KIT alterations (previously known to be recurrent alterations in AM) and CDK4/CCND1 copy number amplifications were identified in the patients. Genetic and transcriptomic alterations in patients with AM were functionally converge to the hallmarks of cancer and oncogenic pathways, including 'proliferative signal persistence', 'apoptotic resistance', and 'activation of invasion and metastasis', despite the heterogeneous somatic mutation profiles of Korean patients with AM. This study may provide a molecular understanding for therapeutic strategy for AM.

Anti-ovarian cancer actions and pharmacological targets of plumbagin

Ovarian cancer is a gynecological malignancy characterized with increasing death rate in the world. It is clinically reported that chemotherapy against ovarian cancer is still found with poor curative effect and potential side effect. Plumbagin is an emerging anti-cancer compound. Although some experimental findings of plumbagin anti-ovarian cancer activity are described, the pharmacological targets should be further explored. In this study, we aimed to investigate the underlying pharmacological activities and targets of plumbagin against ovarian cancer in vitro. As results, in silico docking analysis suggested plumbagin potently treating ovarian cancer through regulating pharmacological targets, including octamer-binding transcription factor 4 (OCT4) and Kruppel-like factor 4 (KLF4). The preliminary experimental data showed that plumbagin treatment inhibited cell growth and induced apoptosis in cancer cells. In addition, decreased mRNA expressions of intracellular OCT4, PCNA, and elevated KLF4 mRNA activation were detected in plumbagin-treated cancer cells. Furthermore, immunostaining determination showed reduced OCT4-positive cells and increased KLF4-positive cells were observed following plumbagin treatments. To sum up, our current findings have preliminarily showed the anti-ovarian cancer benefits of plumbagin, and the pharmacological targets may be identified as KLF4 and OCT4 pathway. Thus, we conclude that plumbagin may be a bioactive compound for ovarian cancer treatment.

Development of a ferroptosis-based model to predict prognosis, tumor microenvironment, and drug response for lung adenocarcinoma with weighted genes co-expression network analysis

Objective: The goal of this study was to create a risk model based on the ferroptosis gene set that affects lung adenocarcinoma (LUAD) patients' prognosis and to investigate the potential underlying mechanisms. Material and Methods: A cohort of 482 LUAD patients from the TCGA database was used to develop the prognostic model. We picked the module genes from the ferroptosis gene set using weighted genes co-expression network analysis (WGCNA). The least absolute shrinkage and selection operator (LASSO) and univariate cox regression were used to screen the hub genes. Finally, the multivariate Cox analysis constructed a risk prediction score model. Three other cohorts of LUAD patients from the GEO database were included to validate the prediction ability of our model. Furthermore, the differentially expressed genes (DEG), immune infiltration, and drug sensitivity were analyzed. Results: An eight-gene-based prognostic model, including PIR, PEBP1, PPP1R13L, CA9, GLS2, DECR1, OTUB1, and YWHAE, was built. The patients from the TCGA database were classified into the high-RS and low-RS groups. The high-RS group was characterized by poor overall survival (OS) and less immune infiltration. Based on clinical traits, we separated the patients into various subgroups, and RS had remarkable prediction performance in each subgroup. The RS distribution analysis demonstrated that the RS was significantly associated with the stage of the LUAD patients. According to the study of immune cell infiltration in both groups, patients in the high-RS group had a lower abundance of immune cells, and less infiltration was associated with worse survival. Besides, we discovered that the high-RS group might not respond well to immune checkpoint inhibitors when we analyzed the gene expression of immune checkpoints. However, drug sensitivity analysis suggested that high-RS groups were more sensitive to common LUAD agents such as Afatinib, Erlotinib, Gefitinib, and Osimertinib. Conclusion: We constructed a novel and reliable ferroptosis-related model for LUAD patients, which was associated with prognosis, immune cell infiltration, and drug sensitivity, aiming to shed new light on the cancer biology and precision medicine.

DrDimont: explainable drug response prediction from differential analysis of multi-omics networks

MOTIVATION

While it has been well established that drugs affect and help patients differently, personalized drug response predictions remain challenging. Solutions based on single omics measurements have been proposed, and networks provide means to incorporate molecular interactions into reasoning. However, how to integrate the wealth of information contained in multiple omics layers still poses a complex problem.

RESULTS

We present DrDimont, Drug response prediction from Differential analysis of multi-omics networks. It allows for comparative conclusions between two conditions and translates them into differential drug response predictions. DrDimont focuses on molecular interactions. It establishes condition-specific networks from correlation within an omics layer that are then reduced and combined into heterogeneous, multi-omics molecular networks. A novel semi-local, path-based integration step ensures integrative conclusions. Differential predictions are derived from comparing the condition-specific integrated networks. DrDimont's predictions are explainable, i.e. molecular differences that are the source of high differential drug scores can be retrieved. We predict differential drug response in breast cancer using transcriptomics, proteomics, phosphosite and metabolomics measurements and contrast estrogen receptor positive and receptor negative patients. DrDimont performs better than drug prediction based on differential protein expression or PageRank when evaluating it on ground truth data from cancer cell lines. We find proteomic and phosphosite layers to carry most information for distinguishing drug response.

AVAILABILITY AND IMPLEMENTATION

DrDimont is available on CRAN: https://cran.r-project.org/package=DrDimont.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.