Chengqi Xu

Weill Cornell PhD of Computational Biomedicine

[Adam C Palmer @ac_palmer]

Most ‘synergistic’ cancer drug combinations fail in the clinic.

Why?

A new paper revisits 100 years of drug combination research to understand the matters key to clinical success
https://t.co/pcWjCxrtbb
1/8

[WCM Englander Institute for Precision Medicine @WCMEnglanderIPM]

Our next @WCMEnglanderIPM #PrecisionMedicine #AI Working Group will feature a talk by Dr. Chengqi Xu, "Tahoe-100M: A Giga-Scale Single-Cell Perturbation Atlas for Context-Dependent Gene Function and Cellular Modeling," on Friday, March 21, 2025 @ 11AM!

[Olivier Elemento @ElementoLab]

AI has thrived on repurposing data not originally designed for it. The next frontier? Data purpose-built for AI, unlocking a new era of innovation

[Chengqi Xu @ChengqiXu]

Kudos to senior authors @WCMEnglanderIPM @ElementoLab @dragon_heng

[Chengqi Xu @ChengqiXu]

Excited to share our new preprint DDI-GPT. 🚀 an AI model blending cutting-edge large language models & knowledge graphs to predict and explain drug-drug interactions like never before -💊🩺 empowering clinicians, elevate patient safety with smarter, safer prescribing decisions.

[bioRxiv Bioinfo @biorxiv_bioinfo]

DDI-GPT: Explainable Prediction of Drug-Drug Interactions using Large Language Models enhanced with Knowledge Graphs https://t.co/c3PQ8vtBbX #biorxiv_bioinfo

[Biology+AI Daily @BiologyAIDaily]

DDI-GPT: Explainable Prediction of Drug-Drug Interactions using Large Language Models enhanced with Knowledge Graphs

• Introducing DDI-GPT, a cutting-edge framework that combines knowledge graphs and large language models (LLMs) to predict and explain drug-drug interactions (DDIs), addressing critical challenges in drug safety.

• DDI-GPT achieves state-of-the-art accuracy with an AUROC of 0.964, outperforming other methods like CASTER by 14% in zero-shot prediction on newly curated FDA datasets.

• The model integrates structured biomedical knowledge from curated graphs into LLMs, enabling rich, context-aware predictions. Its novel “sentence tree” method preserves the relational topology of input data for high-quality results.

• Beyond predictions, DDI-GPT provides interpretability through gene importance scoring and pathway enrichment analysis, offering mechanistic insights into interactions like those involving Bruton’s tyrosine kinase (BTK) inhibitors.

• Validated with independent datasets, DDI-GPT effectively identifies DDIs in real-world cases, demonstrating robustness in predicting interactions for understudied or new drugs.

• The framework supports clinical decision-making with an interactive tool for dynamic DDI analysis, including network visualization of protein-protein interactions and adverse reaction linkages.

• DDI-GPT exemplifies the future of drug safety research by bridging data-driven predictions with explainable insights, essential for precision medicine.

@ElementoLab @ChengqiXu @dragon_heng
💻Code: https://t.co/1sSwSoUnvV
📜Paper: https://t.co/jb9EdDW4bM

#DrugInteractions #Bioinformatics #KnowledgeGraphs #MachineLearning #DrugSafety

[Chengqi Xu @ChengqiXu]

Kudos to senior authors @WCMEnglanderIPM @ElementoLab Heng Pan, Krishna Bulusu

[Chengqi Xu @ChengqiXu]

Check out our new preprint PAIRWISE. A great collaboration with AstraZeneca. We extensively benchmark PAIRWISE with extant models which shows state-of-art performance across tissues and on external DLBCL set. We successfully validated acalabrutinib combinations in HTS experiment.

[bioRxiv Bioinfo @biorxiv_bioinfo]

PAIRWISE: Deep Learning-based Prediction of Effective Personalized Drug Combinations in Cancer https://t.co/wKo16Y0Iae #biorxiv_bioinfo

[Biology+AI Daily @BiologyAIDaily]

PAIRWISE: Deep Learning-based Prediction of Effective Personalized Drug Combinations in Cancer

1. Introducing PAIRWISE: This deep learning model predicts effective, personalized cancer drug combinations by analyzing chemical structures, drug targets, and tumor-specific transcriptome data, directly addressing the challenge of individual tumor heterogeneity.

2. Novel multimodal architecture: PAIRWISE integrates drug chemical structures, drug-target interactions, and tumor transcriptomes using attention mechanisms, enabling it to accurately predict synergistic drug pairs with high AUROC scores, significantly outperforming other models.

3. Enhanced generalizability with transfer learning: PAIRWISE is pre-trained on extensive drug and transcriptome datasets, learning robust molecular representations that allow it to predict synergy for previously unseen drugs, combinations, and cell lines across diverse cancer types.

4. Validation with real-world benchmarks: PAIRWISE demonstrated state-of-the-art performance on an independent dataset, accurately predicting synergistic drug pairs with a 0.72 AUROC in a challenging setting involving Bruton Tyrosine Kinase inhibitors in Diffuse Large B Cell Lymphoma (DLBCL).

5. Personalized therapy predictions: PAIRWISE identified novel drug combinations involving DNA damage response (DDR) pathway inhibitors for aggressive DLBCL, showcasing its potential for tailoring therapies to specific cancer subtypes.

6. Clinical implications: In a cohort of DLBCL patients, PAIRWISE predictions stratified patients into groups with significantly different progression-free survival outcomes, providing a new avenue for guiding combination therapy choices in clinical settings.

7. Impact on precision oncology: PAIRWISE represents a promising advancement for precision oncology by enabling data-driven predictions of personalized drug combinations, potentially reducing the need for high-throughput experimental screening.

@ElementoLab @dragon_heng @bioinpharmatics @marta_milo @ChengqiXu
📜Paper: https://t.co/J0nObO9aad

#CancerResearch #DrugDiscovery #DeepLearning #PrecisionOncology #MolecularBiology

[WCM Englander Institute for Precision Medicine @WCMEnglanderIPM]

Our @WCMEnglanderIPM Drs. @ChengqiXu and Olivier Elemento (@ElementoLab) explore how #LLMs are now being applied to #molecular #biosciences in a new BioChemist paper, "The potential and pitfalls of large language models in molecular biosciences."

https://t.co/wOXQx1MUkG

[WCM Englander Institute for Precision Medicine @WCMEnglanderIPM]

Congratulations to @WCMEnglanderIPM Director Olivier Elemento, Ph.D. (@ElementoLab), below left, and Lorenzo Galluzzi, Ph.D. (@deadoc80) on being named 2023 @Clarivate Highly Cited Researchers!

https://t.co/5tYs1DvDIU

[Fei Wang @feiwang03]

Absolutely thrilled about this exciting endeavor of leading @WCM_AIDH, which will be a vehicle to promote research, facilitate collaborations and democratize techniques of AI for digital health. Thanks for the support from @RainuKaushal @jyotipathak and @WeillCornell leaderships

[Qian Yang @fabulousQian]

I will be recruiting new PhD students who are interested in designing large-scale AI systems (e.g., large language models, pervasive sensing and AI) to improve human experience. Please help retweet🎯 and send recommendations💡!

[Junbum (June) Kim @junbum_kim]

Happy to share that my first first author paper is publicly available! Special thanks to @afrendeiro and @ElementoLab for advising the work!
https://t.co/EqmqwbiuiH
@naturemethods

[Stanford University @Stanford]

"I understand the gravity of being a woman and now a #Nobel laureate in the sciences. There aren't that many of us—yet."

Prof. @CarolynBertozzi on chemistry, mentorship and representation ⬇️

[Volastra Therapeutics @VolastraTx]

When science is art: Check out this image of a colon cancer #organoid captured in our lab.

We’re proud to partner with @WeillCornell to leverage organoids to study tumor biology in vitro, resulting in significantly improved models of patients’ tumors. #SciArt

[Chengqi Xu @ChengqiXu]

Come and Join us🧬

[Felicia New, PhD @FeliciaNNew]

The 2020 Microbiome #Hackathon will take place March 13-15! If you are an undergrad or grad student in Ithaca, please join us for a fun weekend of data science and microbiomes! More info and registration at https://t.co/lZCKduXo6U 🧬💻🦠

#MicrobiomeHack2020

[Chengqi Xu @ChengqiXu]

#SenescenceUK19 I’ve been inspired and learned so much at senescence UK symposium.