AI for Drug Suitability
Overview
Challenge:
Cancer remains one of the most heterogeneous diseases, with patients showing diverse molecular and phenotypic responses to the same therapy. This variability makes it difficult to identify effective drug targets and predict treatment outcomes across individuals.
Why it matters: A “one-size-fits-all” approach to cancer treatment leads to suboptimal responses and drug resistance. Understanding patient-specific molecular profiles and drug–target interactions is essential for tailoring precise therapies and identifying opportunities for drug repurposing—especially when existing compounds could be redirected to novel cancer subtypes.
Current status:
Recent advances in multi-omics profiling and AI-based drug discovery have shown promise in modeling drug–gene interactions. However, these methods often overlook tumor heterogeneity, dynamic signaling rewiring, and population diversity, which are key to predicting individualized drug responses.
Aim:
Develop an AI-driven framework that integrates multi-omics, structural biology, and pharmacogenomic data to model tumor heterogeneity and predict patient-specific drug suitability and repurposing opportunities. By combining deep learning architectures (transformers, GNNs, and multimodal fusion networks), the system will learn cross-scale representations linking mutation profiles, protein structures, and drug mechanisms. This approach will:
· Identify repurposable drugs for resistant or rare cancer subtypes.
· Stratify patients based on molecular signatures and predicted therapeutic response.
· Accelerate precision oncology pipelines by reducing experimental screening costs.
Impact:
This research will enable data-driven, individualized cancer treatment, transforming how therapies are matched to patients and opening new pathways for targeting cancer heterogeneity with repurposed or novel drugs.
Team
Tools
BigQUIC: Big Quadratic Inverse Covariance Estimation
Use Newton’s method, coordinate descent, and METIS clustering to solve the L1 regularized Gaussian MLE inverse covariance matrix estimation problem. https://cran.r-project.org/web/packages/BigQuic/index.html
COUSCOus
Motivation: Current methods for predicting protein residue contacts are valuable but incomplete and do not fully agree. We developed a new method, COUSCOus, that combines advanced statistical techniques to improve accuracy. Our method consistently outperforms the established PSICOV tool across multiple benchmarks and independent tests. This demonstrates that superior statistical approaches can significantly advance protein contact prediction and related fields like gene network analysis.
DeepCrystal
QCRI deep learning models for crystallization propensity prediction, DeepCrystal and BCrystal is ready to compute.
Deepsol
Protein solubility plays a vital role in pharmaceutical research and production yield. For a given protein, the extent of its solubility can represent the quality of its function, and is ultimately defined by its sequence. Thus, it is imperative to develop novel, highly accurate in silico sequence-based protein solubility predictors.
Deployable Code for Early Prediabetes Detection: The PRISQ Model
Leveraging data from the Qatar Biobank, we have created and validated a deployable algorithm for prediabetes screening. The PRISQ model’s code takes basic health metrics as input and outputs a clear risk category (Low, Moderate, High). This allows for seamless integration into digital health platforms, electronic health records, and public screening tools, providing a cost-effective, first-line defense against type 2 diabetes for Middle Eastern populations.
OutSingle
A Python tool for finding outliers in RNA-Seq gene expression count data using SVD/OHT