Biology subjects

Woldring, D.

Publications and source records attributed to Woldring, D..

2 recordsLinked to original sources

bioRxiv2023

AP-LASR: Automated Protein Libraries from Ancestral Sequence Reconstruction

BackgroundAncestral sequence reconstruction (ASR) provides an informative roadmap of evolutionary protein sequence space that benefits protein design and engineering in pursuit of high stability and diverse functionality. Using statistical and biological knowledge, ASR can determine the most probable ancestor among potential alternative amino acid states. However, the inherent uncertainty of ASR can be further leveraged to determine viable "nearby" ancestors with wide-ranging functionalities by sampling alternative amino acid states. ResultsHere we introduce AP-LASR which i) automates ASR and ii) leverages uncertainty in ASR to generate diverse protein sequence libraries that consist of ancestral sequences and near-ancestor sequences. In addition to automating pre-processing tasks (e.g., data cleaning, multiple sequence alignment, and software dependency management), AP-LASR offers several user-definable hyperparameters (e.g., input data size, ancestral probability cut-off, and sequence supplementation) to control the properties of the generated library. AP-LASR features an improved eLnP score (a metric for quantifying reconstructed ancestral sequence confidence) compared to FireProtASR, a well-established ASR workflow, for all four functionally diverse protein families studied. Furthermore, the rigorous statistical analysis undertaken in this study elucidates the influence of hyperparameters on ASR, enabling researchers to refine AP-LASR to their specific research. ConclusionAP-LASR offers an automated ASR experience that surpasses existing software by including a novel library design feature, powering curated protein libraries for wet-lab evaluation. We demonstrate how computational parameters impact the quality of ASR results, library composition, and the tradeoffs therein. AP-LASR offers a powerful tool for protein engineers to efficiently navigate the vast protein sequence landscape. Software available at: https://github.com/WoldringLabMSU/AP-LASR

bioinformatics↗

bioRxiv2023

Protein Fitness Prediction is Impacted by the Interplay of Language Models, Ensemble Learning, and Sampling Methods

Advances in machine learning (ML) and the availability of protein sequences via high-throughput sequencing techniques have transformed our ability to design novel diagnostic and therapeutic proteins. ML allows protein engineers to capture complex trends hidden within protein sequences that would otherwise be difficult to identify in the context of the immense and rugged protein fitness landscape. Despite this potential, there persists a need for guidance during the training and evaluation of ML methods over sequencing data. Two key challenges for training discriminative models and evaluating their performance include handling severely imbalanced datasets (e.g., few high-fitness proteins among an abundance of non-functional proteins) and selecting appropriate protein sequence representations. Here, we present a framework for applying ML over assay-labeled datasets to elucidate the capacity of sampling methods and protein representations to improve model performance in two different datasets with binding affinity and thermal stability prediction tasks. For protein sequence representations, we incorporate two widely used methods (One-Hot encoding, physiochemical encoding) and two language-based methods (next-token prediction, UniRep; masked-token prediction, ESM). Elaboration on performance is provided over protein fitness, length, data size, and sampling methods. In addition, an ensemble of representation methods is generated to discover the contribution of distinct representations to the final prediction score. Within the context of these datasets, the synthetic minority oversampling technique (SMOTE) outperformed undersampling while encoding sequences with One-Hot, UniRep, and ESM representations. In addition, ensemble learning increased the predictive performance of the affinity-based dataset by 4% compared to the best single encoding candidate (F1-score = 97%), while ESM alone was rigorous enough in stability prediction (F1-score = 92%).

bioinformatics↗