| About this Abstract |
| Meeting |
2024 TMS Annual Meeting & Exhibition
|
| Symposium
|
Biological Materials Science
|
| Presentation Title |
Interpretable Data-constrained Machine Learning Model for Predicting the Mechanical Properties of Protein-based Fibers |
| Author(s) |
Akash Pandey, Wei Chen, Sinan Keten |
| On-Site Speaker (Planned) |
Sinan Keten |
| Abstract Scope |
We have developed a machine-learning (ML) pipeline that uses features extracted from just the primary sequence of a protein to predict mechanical properties of its spun fibers. The input features are extracted by collapsing the primary sequence information in a 3-dimensional (3D) representation that captures the frequency and the dynamic property (B-factor) of various pairs of amino acids in the primary sequence. We train and test our proposed pipeline using the experimental data available in the literature for various mechanical properties of the spider dragline silks. Model achieves a R^2 and PCC in the range of 0.6-0.75 and 0.78-0.88 on mechanical property data respectively. Perturbation analysis and position-specific scoring (PSS) is then used to link motifs to properties for guiding fiber design. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Biomaterials, Mechanical Properties, Modeling and Simulation |