Methods & Key Findings :

Metalloproteinases (MPs), a family of zinc-dependent endopeptidases, include matrix metalloproteinases (MMPs) and a disintegrin and metalloproteinases (ADAMs). These enzymes are tightly associated with several diseases when not regulated by their inhibitors and activators, making them attractive targets for developing novel protein-based therapeutics . In this work , we explored application of deep learning models in screening binders to MPs. More specifically, state-of-the-art protein language models (such as ESM and AntiBERTy) were used to extract features from an experimental library of binders and non-binders to three different MPs  (MMP-9, ADAM17, and MMP-3). These models were fine-tuned on the experimental library, and a downstream classifier was trained to separate binders from non-binders. We found that the extracted features were highly effective for predicting binding to MMP-3 and MMP-9 (F1-scores over 90%), but performed less well for ADAM17 (F1-score of 74%). We are currently investigating the reason for this discrepancy and are exploring the use of XAI methods to generate biologically meaningful explanations for the model’s performance and decisions

Collaborators & Students :

• Dr. Maryam Raeeszadeh (UNR) - Faculty of Chemical and Material Engineering at University of Reno at Nevada
• Iftikhar Kalanther (UIS CS Graduate Student), Donald Bleyl (UIS CS Graduate Student), Rushabh Patel (UIS CS Graduate Student), Masoud Kalantar (Graduate Student at UNR)

Publications & Presentations :

• Kalantar et al. (in press). Engineering scFv variants targeting ADAM17. Biomolecules.
• Kalantar et al. (2024). Determining key residues of scFv variants for MMP-9. Comput. Struct. Biotechnol. J., 23, 3759–3770.
• Buxton et al. (2021, Dec). Protein language models for MP inhibitors. IEEE CSCI Conference.