Reproduce “GRAM-DTI: adaptive multimodal representation learning for drug target interaction prediction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Feng Jiang, Amina Mollaysa, Hehuan Ma, Tommaso Mansi, Junzhou Huang, Mangal Prakash, Rui Liao
Researchers from the University of Texas at Arlington and Johnson & Johnson Innovative Medicine developed GRAM-DTI, a framework for drug-target interaction (DTI) prediction that uses adaptive multimodal representation learning. It integrates four data modalities via volume-based contrastive learning, adaptive modality dropout, and IC50 weak supervision, achieving superior performance over baselines on DTI and Mechanism of Action tasks, particularly notable in cold start conditions.
Key insights
- Geometric alignment of four data modalities through Gramian volume-based contrastive learning effectively captures complex, higher-order interdependencies, which single-modality or pairwise methods miss.
- An adaptive modality dropout mechanism, informed by gradient contributions, dynamically adjusts the influence of each modality, preventing over-reliance and enhancing model robustness and generalization.
- Leveraging sparse IC50 activity measurements as weak supervision provides valuable biological context, improving the meaningfulness of learned representations for DTI and MoA prediction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.