CuMMI: AI Model Predicts Nanomaterial-Protein Interactions
A team of researchers has created CuMMI (curriculum-guided multimodal interaction model), an AI model capable of generalizing predictions for nanomaterial-protein interactions (NPI). This model leverages a self-assembled dataset containing millions of NPI instances and employs a multi-stage curriculum focused initially on human plasma, gradually expanding to include a wider range of biofluids. By incorporating 37 features related to protein sequence, structure, and text-encoded experimental context, CuMMI effectively gathers crucial material-specific, biochemical, and environmental insights. The objective of this model is to enhance the understanding of mechanisms and facilitate the rational design of nanomaterials for use in therapeutic and diagnostic fields.
Key facts
- CuMMI is a generalizable, explainable, and transferable model for NPI prediction.
- The model uses a self-constructed million-scale NPI dataset.
- It adopts a multi-stage curriculum centered on human plasma.
- The curriculum includes progressively broader biofluid exposure.
- CuMMI integrates protein sequence, structure, and text-encoded experimental context of 37 features.
- The model captures material-specific, biochemical, and environmental information.
- Sample-level quality weights are assigned to enhance data coverage and generalizability.
- The research is published on arXiv (2507.14245).
Entities
Institutions
- arXiv