COMPASS
Generalizable AI model predicting cancer immunotherapy response across cancer types and treatments. Integrates multi-modal clinical and genomic features validated across multi-center cohorts. Published in Nature Medicine (2026), in collaboration with Harvard Medical School and Roche pRED.
MolMap
Transforms molecular descriptors and fingerprints into structured 2D feature maps, enabling CNNs to learn rich spatial patterns for drug property prediction. Integrated into ZairaChem's automated drug discovery pipeline. Published in Nature Machine Intelligence (2021).
AggMap
Unsupervised multi-omics feature structurization and deep learning toolbox. Organizes high-dimensional omics features into interpretable 2D maps via manifold learning, enabling explainable CNN-based disease diagnosis and biomarker discovery. Published in Nucleic Acids Research (2022). Featured in Cell Press Patterns.
FPPool
A knowledge-guided graph pooling framework that bridges molecular fingerprints and GNNs. Uses chemical fingerprints to decompose molecular graphs into biochemically meaningful substructures, achieving state-of-the-art performance on 40 benchmarks with built-in interpretability. ChemRxiv Preprint (March 2026).
Leadmaster
Automated drug design platform enabling a closed loop of design–synthesis–testing–analysis for small molecule lead compounds. Developed in collaboration with NUS synthetic chemistry lab. Accelerates hit-to-lead timelines from months to weeks.
