Decrypting natural antibiotics to counter biological threats
Abstract
The main goals of this project are to expand our basic science knowledge of AMP activity in order to detect encrypted AMPs actively used within the human body. Our approach involves creating data sets for ML training, combining them with experimental datasets created in our laboratory and available in the literature, and training ML models for encrypted APM detection and screening. In Base Year 1, we will focus on preliminary AMP detection using physicochemical description of amino acid sequences to train recurrent neural networks as well as traditional ML algorithms. We will also use molecular dynamics (MD) to create the first data set of AMPs described through their dynamics, rather than just their sequence. The first milestone will be the creation of a ML-based screening of the human proteome for AMP detection. In Base Year 2, the dynamic and physicochemical descriptors will be combined to train a new class of ML models, never created before. We will also expand the dynamic descriptors data set and create new ML models for additional screening of candidate encrypted AMPs. The second milestone will be the creation of a dynamicdescriptor- based ML model for AMP detection, and a cleavage site detection model. In Option Year 1, we will expand ML training to include screenings for PK/PD properties using all previously collected data, as well as integrating quantum chemical information on dynamical descriptor datasets, to screen candidate AMPs that do not kill commensal bacteria. Our third milestone will be to combine all screening methodologies and provide a ML pipeline for the selection of encrypted AMPs as new MCMs.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 14, 2022
- Source ID
- HDTRA12110014
Entities
People
- Cesar Fuentenunez
Organizations
- Defense Threat Reduction Agency
- University of Pennsylvania