Research map / Probabilistic and scientific modeling / Molecules and proteins
BiGMINT: Biologically-guided Hierarchical Multimodal Integration for Modeling Multiple Compound Activities in Drug Discovery
Pushpak Pati, Bo Li, Abbas Rayabat Khan, Tomé Albuquerque, Steffen Jaensch, Amina Mollaysa, Walid M. Abdelmoula, Samantha J. Allen, Joke Reumers, Helai P. Mohammad, Scott Oloff, Tommaso Mansi and 3 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace mean pooling with attention-based MIL aggregation of HCI patch features | HCI-only baseline (MTL) · U2OS | mean AUCROC (%): 69.40 -> 71.17 | Improved |
| use multi-task learning instead of multi-label learning | HCI-only baseline (MIL) · U2OS | mean AUCROC (%): 67.95 -> 71.17 | Improved |
| add chemoproteomics-guided cross-attention aggregation of HCI patches (ablation) | BiGMINT · U2OS | mean AUCROC (%): 77.72 -> 78.23 | Improved |
| add chemoproteomics-guided cross-attention aggregation of HCI patches (ablation) | BiGMINT · iNeuron | mean AUCROC (%): 75.80 -> 76.51 | Improved |
| add PPI-derived task-task interaction (TTI) prior (ablation) | BiGMINT · U2OS | mean AUCROC (%): 77.02 -> 78.23 | Improved |
| add PPI-derived task-task interaction (TTI) prior (ablation) | BiGMINT · iNeuron | mean AUCROC (%): 74.94 -> 76.51 | Improved |
| add TTI prior to HCI-only MIL model (ablation) | MIL -> MTL (HCI only) · U2OS | mean AUCROC (%): 71.17 -> 72.09 | Improved |
| add TTI prior (ablation) | BiGMINT · U2OS tasks | tasks improved: 114/170 tasks improved; +12% tasks with AUCROC >=0.8 | Improved |
| use outer-product fusion instead of other fusion operators (ablation) | BiGMINT · U2OS and iNeuron | mean AUCROC | Improved |
| combine HCI and chemoproteomics by concatenation instead of unimodal models | multimodal baseline · U2OS | mean AUCROC (%): 71.17 -> 73.65 | Improved |
| use alignment-based pretraining (CLOOME, MolPhenix) instead of integration | multimodal baselines · U2OS | mean AUCROC (%): 63.49 and 70.17 vs BiGMINT 78.23 | Hurt |
| use BiGMINT instead of optimistic per-task best-modality selection (MM-Union) | U2OS/iNeuron · tasks with AUCROC >= 0.8 | number of tasks covered: +56% (U2OS), +5% (iNeuron) | Improved |
Closest papers by meaning
- Monroe: A Molecular Foundation Model for In-context Probabilistic Inference (NeurIPS 2026 · similarity 0.86)
- MF-LAL: Drug Compound Generation Using Multi-Fidelity Latent Space Active Learning (ICML 2025 · similarity 0.86)
- Knowledge Enhanced Representation Learning for Drug Discovery (AAAI 2024 · similarity 0.84)
- Data-Driven Discovery of Dynamical Systems in Pharmacology using Large Language Models (NeurIPS 2024 · similarity 0.84)
- ScreenShot: A Foundation Model for Few-Shot Combination Drug Screening (NeurIPS 2026 · similarity 0.84)
- WelQrate: Defining the Gold Standard in Small Molecule Drug Discovery Benchmarking (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
