Search papers, labs, and topics across Lattice.
This study explores the use of Multitask Bayesian Neural Networks for the simultaneous engineering of multiple protein properties, addressing the limitations of existing machine learning approaches that treat properties in isolation. By evaluating 2,592 models across various architectures and sequence representations, the authors found that Bayesian Last Layer models significantly outperformed others in accuracy, generalization, and calibration, excelling on 70% of the benchmark datasets. Additionally, dimensionality reduction techniques enhanced predictive performance by up to 42%, highlighting the effectiveness of even simple encoding methods in specific contexts.
Bayesian Last Layer models can dramatically improve the accuracy and reliability of multiparameter protein engineering, outperforming traditional methods in 70% of cases.
Simultaneously engineering multiple protein properties remains a major challenge. Existing machine learning-based pipelines for protein engineering often model properties separately, failing to capture their dependencies and trade-offs. Here, we systematically evaluate how Bayesian parameterization on Multitask Neural Networks can enable robust simultaneous protein engineering under scarce, noisy experimental data. We curated a comprehensive set of 27 multiparameter protein datasets. Then, we compared three algorithm architectures spanning low to full Bayesian parameterization across 16 sequence representations and dimensionality reduction (2,592 models). Bayesian Last Layer models delivered the strongest overall accuracy, generalization, and calibration, ranking as the top-performing model on 70% of benchmark datasets. Dimensionality reduction improved predictive performance by up to 42% and enhanced calibration up to 57% across architectures. Notably, simple One-Hot encoding achieved top performance on 25% of benchmark datasets, particularly with larger datasets. These results establish practical design principles for reliable and data-efficient multiparameter protein engineering.
