Roshan Balaji
PhD Scholar, IIT Madras, India
303, Block 2,
Department of Biotechnology,
IIT Madras, Chennai - 600036
I am a PhD scholar and Prime Minister’s Research Fellow at the Indian Institute of Technology Madras (IIT Madras), affiliated with the Wadhwani School of Data Science and the Department of Biotechnology. I am a member of the BioSystems Engineering and Control group, advised by Prof. Nirav Pravinbhai Bhatt.
My research is on machine learning for biology and chemistry: multimodal AI, molecular language models, interpretable molecular representations, and rigorous benchmarks for biological tasks. Recent work includes Leak-CURBER, a leakage-controlled benchmark for enzymatic reaction tasks (NeurIPS 2026 Evaluations and Datasets Track), BiCA, a biomedical dense retriever (AAAI 2026, oral), functional-group-based interpretable property prediction (JCIM 2026), and NovoMolGen, a study of molecular language model pretraining done with collaborators at Mila. In 2025 I was a research intern with Prof. Sarath Chandar at Mila - Quebec AI Institute, working on goal-aware retrosynthesis planning.
I hold a BTech in Biotechnology from NIT Calicut, where I graduated as a gold medalist. I also contribute to the OpenFold open-source initiative, co-organize the GCLR workshop at AAAI, and enjoy teaching. I have served as a teaching assistant for several courses at IIT Madras.
News
| Sep 25, 2026 | Our paper Leak-CURBER: A Leakage-Controlled Multimodal Evaluation Benchmark for Enzymatic Reaction Tasks (equal first author) was accepted at the NeurIPS 2026 Evaluations and Datasets Track  |
|---|---|
| May 11, 2026 | Our paper Functional Groups are All you Need for Chemically Interpretable Molecular Property Prediction is published in the Journal of Chemical Information and Modeling |
| Jan 20, 2026 | Organized the GCLR Workshop at AAAI 2026 as Organizer and Program Chair  |
| Nov 20, 2025 | Our paper BiCA: Effective Biomedical Dense Retrieval with Citation-Aware Hard Negatives was accepted as an oral (top 5%) at AAAI 2026 |
| Nov 15, 2025 | I am now contributing to the OpenFold open-source initiative  |
Latest Posts
| Nov 16, 2020 | Protein-Ligand Binding Site Prediction |
|---|
Selected Publications
2026
- NeurIPSLeak-CURBER: A Leakage-Controlled Multimodal Evaluation Benchmark for Enzymatic Reaction TasksIn The Fortieth Annual Conference on Neural Information Processing Systems (NeurIPS), Evaluations and Datasets Track, 2026* Equal contribution
- JCIMFunctional Groups are All you Need for Chemically Interpretable Molecular Property PredictionJournal of Chemical Information and Modeling, 2026
- AAAIBiCA: Effective Biomedical Dense Retrieval with Citation-Aware Hard NegativesIn Proceedings of the AAAI Conference on Artificial Intelligence, 2026Oral presentation (top 5%)