Biology subjects

Sundar, V.

Publications and source records attributed to Sundar, V..

4 recordsLinked to original sources

bioRxiv2024

FLIGHTED: Inferring Fitness Landscapes from Noisy High-Throughput Experimental Data

Machine learning (ML) for protein design requires large protein fitness datasets generated by high-throughput experiments for training, fine-tuning, and benchmarking models. However, most models do not account for experimental noise inherent in these datasets, harming model performance and changing model rankings in benchmarking studies. Here we develop FLIGHTED, a Bayesian method of accounting for uncertainty by generating probabilistic fitness landscapes from noisy high-throughput experiments. We demonstrate how FLIGHTED can improve model performance on two categories of experiments: single-step selection assays, such as phage display and SELEX, and a novel high-throughput assay called DHARMA that ties activity to base editing. We then compare the performance of standard machine-learning models on fitness landscapes generated with and without FLIGHTED. Accounting for noise significantly improves model performance, especially of CNN architectures, and changes relative rankings on numerous common benchmarks. Based on our new benchmarking with FLIGHTED, data size, not model scale, currently appears to be limiting the performance of protein fitness models, and the choice of top model architecture matters more than the protein language model embedding. Collectively, our results indicate that FLIGHTED can be applied to any high-throughput assay and any machine learning model, making it straightforward for protein designers to account for experimental noise when modeling protein fitness.

bioinformatics↗

bioRxiv2022

Neural Network-Derived Potts Models for Structure-Based Protein Design using Backbone Atomic Coordinates and Tertiary Motifs

AO_SCPLOWBSTRACTC_SCPLOWDesigning novel proteins to perform desired functions, such as binding or catalysis, is a major goal in synthetic biology. A variety of computational approaches can aid in this task. An energy-based framework rooted in the sequence-structure statistics of tertiary motifs (TERMs) can be used for sequence design on pre-defined backbones. Neural network models that use backbone coordinate-derived features provide another way to design new proteins. In this work, we combine the two methods to make neural structure-based models more suitable for protein design. Specifically, we supplement backbone-coordinate features with TERM-derived data, as inputs, and we generate energy functions as outputs. We present two architectures that generate Potts models over the sequence space: TERMinator, which uses both TERM-based and coordinate-based information, and COORDinator, which uses only coordinate-based information. Using these two models, we demonstrate that TERMs can be utilized to improve native sequence recovery performance of neural models. Furthermore, we demonstrate that sequences designed by TERMinator are predicted to fold to their target structures by AlphaFold. Finally, we show that both TERMinator and COORDinator learn notions of energetics, and these methods can be fine-tuned on experimental data to improve predictions. Our results suggest that using TERM-based and coordinate-based features together may be beneficial for protein design and that structure-based neural models that produce Potts energy tables have utility for flexible applications in protein science. CodeCode will be made publically available at https://github.com/alexjli/terminator_public

biophysics↗

bioRxiv2020

Using Single Protein/Ligand Binding Models to Predict Active Ligands for Unseen Proteins

Machine learning models that predict which small molecule ligands bind a single protein target report high levels of accuracy for held-out test data. An important challenge is to extrapolate and make accurate predictions for new protein targets. Improvements in drug-target interaction (DTI) models that address this challenge would have significant impact on drug discovery by eliminating the need for high-throughput screening experiments against new protein targets. Here we propose a data augmentation strategy that addresses this challenge to enable accurate prediction in cases where no experimental data is available. To proceed, we first build single protein-ligand binding models and use these models to predict whether additional ligands bind to each protein. We then use these predictions to augment the experimental data, train standard DTI models, and predict interactions between unseen test proteins and ligands. This approach achieves Area Under the Receiver Operator Characteristic (AUC) > 0.9 consistently on test sets consisting exclusively of proteins and ligands for which the model is given no experimental data. We verify that performance improvements extend to held-out test proteins distant from the training set. Our data augmentation framework can be applied to any DTI model, and enhances performance on a range of simple models.

bioinformatics↗

bioRxiv2018

MODE for detecting and estimating genetic causal variants

Determining the genetic causal variants and estimating their effect sizes are considered to be correlated but independent problems. Fine-mapping studies often rely on the ability to integrate useful functional annotation information into genome wide association univariate/multivariate analysis. In the present study, by modeling the probability of a SNP being causal and its effect size as a set of correlated Gaussian/non-Gaussian random variables, we design an optimization routine for simultaneous fine-mapping and effect size estimation. The algorithm is released as an open source C package MODE.\n\nAvailability and Implementation: http://sites.google.com/site/sundarvelkur/mode\n\nContact: amdale@ucsd.edu, svelkur@ucsd.edu

bioinformatics↗