Ashley Babjac

“It is a capitol mistake to theorize before one has data.”
— Sir Arthur Conan Doyle, Author of Sherlock Holmes

Current Research

Custom BERT Architectures: Using LSTMs and BERT transformer architectures to model RNA sequences for downstream classification (high/low or high/medium/low). Application of SHAP and attention for explainable AI and robust CUB calculation.
Protein Language Modeling: Applications of BERT and ESM for identifying characteristics of deep vs surface organisms of the same species. Will eventually modify for translation/next word prediction.
Feature Importance Pipelines: Creating robust feature importance pipelines using LASSO, bootstrapping and permutation importance to detect important features given any biological dataset.
Genomics and CNNs: Using GradCAM and CNNs for determining functional region of a protein based on graphlet sketches of 3D protein structures.
Explainable Deep Learning: Building model pipelines for UT Medical pre-term birth. Emphasis on interpretation of features and accurate prediction across a variety of tasks.
Vision and Clustering: Defining less manual cluster labeling in single-cell-omics using visualization technology (with UTK SeeLab).

Publications

Babjac, A., Royalty, T., Steen, A.D. and Emrich, S.J., 2022, August. A Comparison of Dimensionality Reduction Methods for Large Biological Data. In Proceedings of the 13th ACM International Conference on Bioinformatics, Computational Biology and Health Informatics (pp. 1-7).
Babjac, A., Li, J. and Emrich, S., 2021, December. Fine-Grained Synonymous Codon Usage Patterns and their Potential Role in Functional Protein Production. In 2021 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) (pp. 2187-2193). IEEE.
*Oduwole, I. and *Babjac, A., Steen, A.D., Lloyd, K., and Emrich, S.J. (et. al.) (2023 likely). Understanding Cultured and Uncultured Patterns in GEM Microbiomes using LASSO Feature Selection. Journal submission likely early 2023. (In progress).
DREAM Challenge (2022) – Predicting gene expression using millions of random promoter sequences. (UTKBioinformatics on leaderboard for co-authorship). Pre-print: https://www.biorxiv.org/content/10.1101/2023.04.26.538471v1
DREAM Challenge (2022) – Preterm birth prediction (microbiome). (UTKBioinformatics on leaderboard for co-authorship). Pre-print: https://www.medrxiv.org/content/10.1101/2023.03.07.23286920v2
DREAM Challenge (2023) – FINRISK heart failure time-to-event prediction (microbiome). (UTKBioinformatics on leaderboard for co-authorship).

Education

Fellowships

Awards