Wang, Liewei

Our research uncovers a role for ATR in responding to ECM stiffness and promoting epithelial-to-mesenchymal transition (EMT) and metastasis. ATR, when deubiquitinated and upregulated by USP21 under enhanced ECM stiffness conditions, phosphorylates the nuclear protein SUN2, which promotes β-catenin nuclear translocation and EMT. ATM-mediated EMT promotes polymorphonuclear myeloid-derived suppressor cell recruitment and inhibits CD103 + dendritic cells, fostering an immunosuppressive tumor milieu. ATR inhibition disrupts this malignant cascade by promoting mesenchymal-to-epithelial transition to enhance antitumor immunity and mitigate metastases. Consistently, circulating HLA-DR + dendritic cells were also enhanced following treatment with the ATR inhibitor berzosertib in patients with therapeutically resistant early-stage breast cancer. Our data suggest that ATR-targeted therapy may be optimized by considering both DNA damage–dependent and EMT-inducing effects of ATR ATR promotes metastasis and immunosuppression through EMT process

Abstract Background Electronic health record (EHR)–based prognostic modeling is increasingly used in oncology, yet incorporating pharmacogenomic (PGx) knowledge derived from experimental systems into clinical prediction frameworks remains challenging. This gap is driven by fundamental mismatches between controlled drug–mutation assays and heterogeneous, incomplete real-world clinical data. Methods We propose a representation transfer framework that integrates PGx embeddings learned from large-scale in vitro pharmacogenomic screens into patient-level EHR models. A frozen pharmacogenomic encoder is used to generate interaction-aware embeddings from patient mutation profiles and administered therapies, which are aggregated into a fixed-length PGx Complementarity Representation. These representations are incorporated into multimodal survival prediction models alongside standard clinical features. Performance was evaluated using systematic modality ablation analyses, attribution analyses, and exploratory unsupervised representation analyses. Results Integrating PGx embeddings yielded consistent performance improvements across all evaluated modality combinations. Relative gains were largest in modality-sparse settings, where baseline EHR features encode limited biological context, and were attenuated—but remained significant—in biologically enriched configurations. Attribution analyses indicated that PGx embeddings contributed non-redundant predictive signal beyond standard clinical features. Exploratory unsupervised analyses further demonstrated that the learned representations exhibit interpretable association patterns aligned with known therapeutic exposures and pathway-level associations. Conclusion These findings suggest that externally learned pharmacogenomic representations can be transferred into real-world EHR models as a context-dependent, non-redundant augmentation. By framing PGx knowledge as an interaction-aware representation rather than a mechanistic model, this work provides an informatics framework for integrating experimental pharmacogenomic data into clinical prediction tasks in a reproducible and interpretable manner.

ABSTRACT Major depressive disorder (MDD) is a substantial public health challenge. Pharmacogenomics (PGx), which identifies genetic variations that predict drug treatment outcomes, may have utility for clinical practice, but adequate representation of all populations is needed. As precision medicine in psychiatry moves towards the use of Artificial Intelligence (AI) and Machine Learning (ML) to predict treatment outcomes using PGx data, representation of diverse populations will be especially important in order to mitigate algorithmic bias and achieve equitable and generalizable findings. This work sought to quantify population diversity in pharmacogenomic studies of MDD through a systematic review. Data from 390 MDD antidepressant PGx studies were extracted from 5739 articles screened. Studies summarized were predominantly conducted in Europe, East Asia, and North America. Across all global studies, the study population comprised 57.3% White, 36.4% Asian, 1.7% Black, 3.5% Hispanic/Latino, and 0.1% Native American or Indigenous participants. Only sixty‐three (16.2%) studies included Black or Latino/Hispanic patients. Additionally, Black, Asian, Hispanic/Latino, and American Indian/Alaska Native populations were statistically underrepresented in U.S. study populations when compared to national census data, while Asian and Black populations were underrepresented in the United Kingdom. The overrepresentation of participants from a limited number of countries combined with the underrepresentation of Black and Hispanic/Latino populations could impact the extent to which pharmacogenomic testing and associated AI/ML‐based PGx tools could individualize antidepressant medication regimens for treating MDD.

Triple-negative breast cancer (TNBC) represents the most malignant subtype of breast cancer. The clinical application of PARP inhibitors (PARPi) is limited by the low frequency of BRCA1/2 mutations in TNBC. Here, we identified that MTAP deletion sensitized genotoxic agents in our clinical cohort of metastatic TNBC. Further study demonstrated that MTAP deficiency or inhibition rendered TNBC susceptibility to chemotherapeutic agents, particularly PARPi. Mechanistically, targeting MTAP that synergized with PARPi by disrupting the METTL16-MAT2A axis involved in methionine metabolism and depleting in vivo s-adenosylmethionine (SAM) levels. Exhausted SAM in turn impaired PARPi-induced DNA damage repair through attenuation of MRE11 recruitment and end resection by diminishing MRE11 methylation. Notably, brain metastatic TNBC markedly benefited from a lower dose of PARPi and MTAP deficiency/inhibition synergy due to the inherently limited methionine environment in the brain. Collectively, our findings revealed a feed-forward loop between methionine metabolism and DNA repair through SAM, highlighting a therapeutic strategy of PARPi combined with MTAP deficiency/inhibition for TNBC.

ABSTRACTPharmacogenetics is a promising strategy to facilitate individualized care for patients with Major Depressive Disorder (MDD). Research is ongoing to identify the optimal genetic markers for predicting outcomes to antidepressant therapies. The primary aim of this systematic review was to summarize antidepressant pharmacogenetic studies to enhance understanding of the genes, variants, datatypes/methodologies, and outcomes investigated in the context of MDD. The secondary aim was to identify clinical genetic panels indicated for antidepressant prescribing and summarize their genes and variants. Screening of N = 5793 articles yielded N = 390 for inclusion, largely comprising adult (≥ 18 years) populations. Top‐studied variants identified in the search were discussed and compared with those represented on the N = 34 clinical genetic panels that were identified. Summarization of articles revealed sources of heterogeneity across studies and low rates of replicability of pharmacogenetic associations. Heterogeneity was present in outcome definitions, treatment regimens, and differential inclusion of mediating variables in analyses. Efficacy outcomes (i.e., response, remission) were studied at greater frequency than adverse‐event outcomes. Studies that used advanced analytical approaches, such as machine learning, to integrate variants with complimentary biological datatypes were fewer in number but achieved higher rates of significant associations with treatment outcomes than candidate variant approaches. As large biological datasets become more prevalent, machine learning will be an increasingly valuable tool for parsing the complexity of antidepressant response. This review provides valuable context and considerations surrounding pharmacogenetic associations in MDD which will help inform future research and translation efforts for guiding antidepressant care.

Background: Prostate cancer is the most common cancer in men in the United States with an estimated 288,300 new cases diagnosed in 2023. The treatment strategies include androgen deprivation therapies (ADTs) for metastatic prostate cancer (mPC) but 10-20% of patients eventually progress to castration-resistant prostate cancer (CRPC). However, nearly all patients develop primary and acquired resistance overtime, and developing therapies and identifying novel molecular targets are essential in enhancing the clinical outcomes of patients with prostate cancer. In this study, we report a small molecule that belongs to a class of tricyclic quinolones, QL47 that was previously screened among various other compounds in prostate PDX organoid models and showed a potent anti-proliferative activity. Methods: QL47 was tested on prostate cancer cell lines (22Rv1, 22RV1 abiraterone-resistant and DU145) for its anti-proliferative potential. Further, we used the biotinylated QL47 (QL47B) to pulldown its potential targets and we identified the targets through immunoprecipitation-mass spectrometry (IP-MS) technique followed by Western blot analysis. Cell cycle and cell death analyses were performed after staining the cells with propidium iodide and Annexin V followed by flow cytometry analysis. Oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) was investigated using the Seahorse assay. Results: The initial screening of QL47 in prostate cancer cell lines showed that it inhibits the proliferation at an IC50 of 50-200 nM. Through IP-MS and Western blot analysis, we identified a member of the pyruvate dehydrogenase complex dihydrolipoamide S-acetyltransferase (DLAT) as a direct target of QL47. Mechanistically, we found that QL47 does not increase cell death or affect cell cycle, but inhibits metabolic function, with a decrease in both oxygen consumption rate (OCR) and extracellular acidification rate (ECAR). Conclusion: We have identified a novel mechanism of action of QL47 on the pyruvate dehydrogenase complex, and we plan to further validate the safety and efficacy for the treatment of prostate cancer. Citation Format: Grace Nirmala Jesuthankaraj, August John, Huanyao Gao, Arun Kanakkanthara, Richard Weinshilboum, Liewei Wang. Targeting metabolic dysfunction in prostate cancer: the anti-proliferative effects of ql47 through inhibition of dlat [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 4297.

Abstract Deep learning has advanced digital pathology and multi-omics data analysis. However, there is still a critical need to integrate these modalities to better understand complex diseases. Deep learning models, such as Prov-Gigapath, a state-of-the-art pathology foundation model, have been developed for analyzing whole-slide images (WSI). Our previously published deep-learning framework, OmicsFootPrint (OFP), transforms high-dimensional multi-omics data into intuitive genomically-arranged circular images and utilizes transfer learning to overcome the challenges of high-dimensional, low-sample-size data. In this study, we aimed to combine WSI and multi-omics data to develop a framework capable of distinguishing between invasive ductal carcinoma (IDC) and invasive lobular carcinoma (ILC), which is clinically significant as it may help explain differential responses to therapies, such as tamoxifen versus letrozole. We enhanced the performance of our OFP model by dividing each multi-omics circular image (1024x1024x3 pixels) into 16 equal-sized patches (256x256x3 pixels). Each patch was trained using a separate EfficientNetV2 architecture with pre-trained ImageNet weights. The probability outcomes from these 16 models were subsequently combined through a meta-learner model using the AutoGluon framework. Concurrently, WSI images corresponding to the same patients were independently trained, maintaining consistent train/validation/test splits. The Prov-Gigapath model, which was trained using a dataset from the Providence Health System, was employed to extract patch-level embeddings from the WSI data. These embeddings were fine-tuned using a LongNet-based masked autoencoder to predict outcomes. Finally, predictions from the OFP model and the WSI-based Prov-Gigapath model were integrated through a meta-learner, preserving the original train/validation/test split. Model performances (AUC) were evaluated using the held-out test set. The gold standard used for the analysis was histological identification of IDC versus ILC, against which the performance of omics data alone, WSI data alone, and the multi-modal integration approach were compared. We evaluated our integrated model as a proof-of-concept using The Cancer Genome Atlas (TCGA) breast cancer cohort, comprising 92 ILC and 391 IDC samples. Multi-omics data, including gene expression, microRNA, copy number variation, and RPPA, were analyzed using the OFP framework, while WSI data were processed using the Prov-Gigapath method. The results demonstrated that the multi-modal integration approach significantly outperformed individual modalities, achieving an AUC of 0.98 compared to 0.87 for OFP (original), 0.95 for OFP (16-patches), and 0.92 for WSI. Integrating WSI and multi-omics data improves predictive accuracy, significantly improving the classification of breast cancer subtypes over histology alone. Citation Format: Naresh Prodduturi, Xiaojia Tang, Hamid R. Tizhoosh, Kevin J. Thompson, Richard Weinshilboum, Karthik V. Giridhar, Judy C. Boughey, Eric W. Klee, Liewei Wang, Matthew P. Goetz, Vera Suman, Krishna R. Kalari. A multi-modal learning framework to integrate digital pathology image and multi-omics data in breast cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2429.

Abstract Background: Pharmacogenomic studies on antidepressant treatment outcomes could be conducted using previously collected data from electronic health record (EHR)-linked biobanks. However, absence of EHR based outcome measures is an unmet need in designing such studies We aimed to define EHR-derived antidepressant outcome measures and explore their utility in showing associations between treatment outcomes and Cytochrome P450 (CYP) metabolizer phenotypes in a proof-of-concept study. Methods: Using data from the EHR-linked cohort, Right Drug, Right Dose, Right Time: Using Genomic Data to Individualize Treatment (RIGHT 10K) Study, we collected prescription data and patient health questionnaire 9 (PHQ-9) scores to compute 3 proxy measures for antidepressant response, efficacy, and acceptability: change in PHQ-9 scores, longest treatment interval with a single antidepressant, and antidepressant non-refill. Subsequently, we tested the association of both prescription-based outcomes with DNA-predicted CYP metabolizer phenotypes in European-ancestry participants. Results: We identified 3920 RIGHT 10K participants with at least 1 antidepressant prescription and European-ancestry. Participants had a mean age of 61 years and 72% were women. Implementation of the PHQ-9 outcome was not feasible because of missingness. Of both prescription-based outcomes, antidepressant non-refill reproduced several known antidepressant-CYP interactions. However, the pilot was limited by small subgroups of participants with non-normal metabolizer phenotypes. Conclusions: Derived from structured data, antidepressant non-refill is a promising outcome measure for EHR-linked biobanks that partially reproduced antidepressant-CYP interactions. However, testing on larger datasets is necessary to understand whether it would be a useful for pharmacogenomic research.

Abstract Objective The fast accumulation of vast pharmacogenomics data of cancer cell lines provide unprecedented opportunities for drug sensitivity prediction (DSP), a crucial prerequisite for the advancement of precision oncology. Recently, Generative Large Language Models (LLM) have demonstrated performance and generalization prowess across diverse tasks in the field of natural language processing (NLP). However, the structured format of the pharmacogenomics data poses challenge for the utility of LLM in DSP. Therefore, the objective of this study is multi-fold: to adapt prompt engineering for structured pharmacogenomics data toward optimizing LLM’s DSP performance, to evaluate LLM’s generalization in real-world DSP scenarios, and to compare LLM’s DSP performance against that of state-of-the-science baselines. Methods We systematically investigated the capability of the Generative Pre-trained Transformer (GPT) as a DSP model on four publicly available benchmark pharmacogenomics datasets, which are stratified by five cancer tissue types of cell lines and encompass both oncology and non-oncology drugs. Essentially, the predictive landscape of GPT is assessed for effectiveness on the DSP task via four learning paradigms: zero-shot learning, few-shot learning, fine-tuning and clustering pretrained embeddings. To facilitate GPT in seamlessly processing the structured pharmacogenomics data, domain-specific novel prompt engineering is employed by implementing three prompt templates (i.e., Instruction, Instruction-Prefix, Cloze) and integrating pharmacogenomics-related features into the prompt. We validated GPT’s performance in diverse real-world DSP scenarios: cross-tissue generalization, blind tests, and analyses of drug-pathway associations and top sensitive/resistant cell lines. Furthermore, we conducted a comparative evaluation of GPT against multiple Transformer-based pretrained models and existing DSP baselines. Results Extensive experiments on the pharmacogenomics datasets across the five tissue cohorts demonstrate that fine-tuning GPT yields the best DSP performance (28% F1 increase, p-value= 0.0003) followed by clustering pretrained GPT embeddings (26% F1 increase, p-value= 0.0005), outperforming GPT in-context learning (i.e., few-shot). However, GPT in the zero-shot setting had a big F1 gap, resulting in the worst performance. Within the scope of prompt engineering, performance enhancement was achieved by directly instructing GPT about the DSP task and resorting to a concise context format (i.e., instruction-prefix), leading to F1 performance gain of 22% (p-value=0.02); while incorporation of drug-cell line prompt context derived from genomics and/or molecular features further boosted F1 score by 2%. Compared to state-of-the-science DSP baselines, GPT significantly asserted superior mean F1 performance (16% gain, p-value<0.05) on the GDSC dataset. In the crosstissue analysis, GPT showcased comparable generalizability to the within-tissue performances on the GDSC and PRISM datasets, while statistically significant F1 performance improvements on the CCLE (8%, p-value=0.001) and DrugComb (19%, p-value=0.009) datasets. Evaluation on the challenging blind tests suggests GPT’s competitiveness on the CCLE and DrugComb datasets compared to random splitting. Furthermore, analyses of the drug-pathway associations and log probabilities provided valuable insights that align with previous DSP findings. Conclusion The diverse experiment setups and in-depth analysis underscore the importance of generative LLM, such as GPT, as a viable in silico approach to guide precision oncology. Availability https://github.com/bioIKEA/SensitiveCancerGPT

Abstract The OmicsFootPrint framework addresses the need for advanced multi-omics data analysis methodologies by transforming data into intuitive two-dimensional circular images and facilitating the interpretation of complex diseases. Utilizing deep neural networks and incorporating the SHapley Additive exPlanations algorithm, the framework enhances model interpretability. Tested with The Cancer Genome Atlas data, OmicsFootPrint effectively classified lung and breast cancer subtypes, achieving high area under the curve (AUC) scores—0.98 ± 0.02 for lung cancer subtype differentiation and 0.83 ± 0.07 for breast cancer PAM50 subtypes, and successfully distinguished between invasive lobular and ductal carcinomas in breast cancer, showcasing its robustness. It also demonstrated notable performance in predicting drug responses in cancer cell lines, with a median AUC of 0.74, surpassing nine existing methods. Furthermore, its effectiveness persists even with reduced training sample sizes. OmicsFootPrint marks an enhancement in multi-omics research, offering a novel, efficient and interpretable approach that contributes to a deeper understanding of disease mechanisms.