Biology subjects

Hawkins-Hooker, A.

Publications and source records attributed to Hawkins-Hooker, A..

7 recordsLinked to original sources

bioRxiv2024

PertEval-scFM: Benchmarking Single-Cell Foundation Models for Perturbation Effect Prediction

In silico modeling of transcriptional responses to perturbations is crucial for advancing our understanding of cellular processes and disease mechanisms. We present PertEval-scFM, a standardized framework designed to evaluate models for perturbation effect prediction. We apply PertEval-scFM to benchmark zero-shot single-cell foundation model (scFM) embeddings against base-line models to assess whether these contextualized representations enhance perturbation effect prediction. Our results show that scFM embeddings do not provide consistent improvements over baseline models, especially under distribution shift. Overall, this study provides a systematic evaluation of zero-shot scFM embeddings for perturbation effect prediction, highlighting the challenges of this task and revealing the limitations of current-generation scFMs. Our findings underscore the need for specialized models and high-quality datasets that capture a broader range of cellular states. Source code and documentation can be found at: https://github.com/aaronwtr/PertEval.

bioinformatics↗

bioRxiv2024

Likelihood-based fine-tuning of protein language models for few-shot fitness prediction and design

Machine learning models trained on measurements of protein functional properties are widely used to accelerate laboratory-based protein design campaigns. To maximise the signal that can be extracted from limited experimental data, sequence embeddings produced by protein language models (PLMs) are often used as the basis of supervised fitness predictors. However, embedding-based predictors do not directly exploit the distributional information encoded in PLM likelihoods after self-supervised or generative pretraining on natural protein sequences. In contrast, likelihood-based fine-tuning approaches exploit this prior knowledge by directly updating pretrained PLM likelihoods to reflect observed fitness differences between sequences. While likelihood-based fine-tuning methods have been proposed previously, a conclusive comparison of their performance against state-of-the-art embedding-based methods has been lacking. To address this gap, we conduct a comprehensive empirical evaluation of both fine-tuning strategies on a representative set of protein fitness datasets from the ProteinGym benchmark. To ensure our evaluation is applicable across different PLM classes, we develop a simple, unified framework for likelihood-based fine-tuning that applies to models trained with various objectives. Across model classes and fitness datasets, likelihood-based fine-tuning consistently outperforms embedding-based methods previously reported as state-of-the-art, with the largest gains in low-data settings. Finally, to highlight the practical relevance of these findings, we demonstrate that the best-performing fine-tuning strategies can substantially improve the maximal fitness of designed sequences in multi-round in silico optimisation campaigns.

bioinformatics↗

bioRxiv2023

Chainsaw: protein domain segmentation with fullyconvolutional neural networks

0.1 MotivationProtein domains are fundamental units of protein structure and play a pivotal role in understanding folding, function, evolution, and design. The advent of accurate structure prediction techniques has resulted in an influx of new structural data, making the partitioning of these structures into domains essential for inferring evolutionary relationships and functional classification. 0.2 ResultsThis manuscript presents Chainsaw, a supervised learning approach to domain parsing that achieves accuracy that surpasses current state-of-the-art methods. Chainsaw uses a fully convolutional neural network which is trained to predict the probability that each pair of residues is in the same domain. Domain predictions are then derived from these pairwise predictions using an algorithm that searches for the most likely assignment of residues to domains given the set of pairwise co-membership probabilities. Chainsaw matches CATH domain annotations in 78% of protein domains versus 72% for the next closest method. When predicting on AlphaFold models expert human evaluators were twice as likely to prefer Chainsaws predictions versus the next best method. 0.3 Availability and ImplementationCode implementation of Chainsaw is available at github.com/JudeWells/chainsaw.

molecular biology↗

bioRxiv2022

The ENCODE Imputation Challenge: A critical assessment of methods for cross-cell type imputation of epigenomic profiles

Functional genomics experiments are invaluable for understanding mechanisms of gene regulation. However, comprehensively performing all such experiments, even across a fixed set of sample and assay types, is often infeasible in practice. A promising alternative to performing experiments exhaustively is to, instead, perform a core set of experiments and subsequently use machine learning methods to impute the remaining experiments. However, questions remain as to the quality of the imputations, the best approaches for performing imputations, and even what performance measures meaningfully evaluate performance of such models. In this work, we address these questions by comprehensively analyzing imputations from 23 imputation models submitted to the ENCODE Imputation Challenge. We find that measuring the quality of imputations is significantly more challenging than reported in the literature, and is confounded by three factors: major distributional shifts that arise because of differences in data collection and processing over time, the amount of available data per cell type, and redundancy among performance measures. Our systematic analyses suggest several steps that are necessary, but also simple, for fairly evaluating the performance of such models, as well as promising directions for more robust research in this area.

bioinformatics↗

bioRxiv2022

Getting Personal with Epigenetics: Towards Machine-Learning-Assisted Precision Epigenomics

Epigenetic modifications are dynamic control mechanisms involved in the regulation of gene expression. Unlike the DNA sequence itself, they vary not only between individuals but also between different cell types of the same individual. Exposure to environmental factors, somatic mutations, and ageing contribute to epigenomic changes over time, which may constitute early hallmarks or causal factors of disease. Epigenetic changes are reversible and, therefore, promising therapeutic targets. However, mapping efforts to determine an individuals cell-type-specific epigenome are constrained by experimental costs. We developed eDICE, an attention-based deep learning model, to impute epigenomic tracks. eDICE achieves improved overall performance compared to previous models on the reference Roadmap epigenomes. Furthermore, we present a proof of concept for the imputation of personalised epigenomic measurements on the ENTEx dataset, where eDICE correctly predicts individual- and cell-type-specific epigenetic patterns. This case study constitutes an important step towards robustly employing machine-learning-based approaches for personalised epigenomics.

bioinformatics↗

bioRxiv2020

Generating functional protein variants with variational autoencoders

The design of novel proteins with specified function and controllable biochemical properties is a longstanding goal in bio-engineering with potential applications across medicine and nanotechnology. The vast expansion of protein sequence databases over the last decades provides an opportunity for new approaches which seek to learn the sequence-function relationship directly from natural sequence variation. Advances in deep generative models have led to the successful modelling of diverse kinds of high-dimensional data, from images to molecules, allowing the generation of novel, realistic samples. While deep models trained on protein sequence data have been shown to learn biologically meaningful representations helpful for a variety of downstream tasks, their potential for direct use in protein engineering remains largely unexplored. Here we show that variational autoencoders trained on a dataset of almost 70000 luciferase-like oxidoreductases can be used to generate novel, functional variants of the luxA bacterial luciferase. We propose separate VAE models to work with aligned sequence input (MSA VAE) and raw sequence input (AR-VAE), and offer evidence that while both are able to reproduce patterns of amino acid usage characteristic of the family, the MSA VAE is better able to capture long-distance dependencies reflecting the influence of 3D structure. To validate the practical utility of the models, we used them to generate variants of luxA whose function was tested experimentally. As further evidence of the practicality of these methods for design, we showed that conditional variants of both models could be used to increase the solubility of luxA without disrupting function. Altogether 18/24 of the variants generated using the AR-VAE and 21/23 variants generated using the MSA VAE retained some luminescence activity, despite containing as many as 35 differences relative to any training set sequence. These results demonstrate the feasibility of using deep generative models to explore the space of possible protein sequences and generate useful variants, providing a method complementary to rational design and directed evolution approaches.

bioinformatics↗

bioRxiv2018

Projection layers improve deep learning models of regulatory DNA function

With the increasing application of deep learning methods to the modelling of regulatory DNA sequences has come an interest in exploring what types of architecture are best suited to the domain. Networks designed to predict many functional characteristics of noncoding DNA in a multitask framework have to recognise a large number of motifs and as a result benefit from large numbers of convolutional filters in the first layer. The use of large first layers in turn motivates an exploration of strategies for addressing the sparsity of output and possibility for overfitting that result. To this end we propose the use of a dimensionality-reducing linear projection layer after the initial motif-recognising convolutions. In experiments with a reduced version of the DeepSEA dataset we find that inserting this layer in combination with dropout into convolutional and convolutional-recurrent architectures can improve predictive performance across a range of first layer sizes. We further validate our approach by incorporating the projection layer into a new convolutional-recurrent architecture which achieves state of the art performance on the full DeepSEA dataset. Analysis of the learned projection weights shows that the inclusion of this layer simplifies the networks internal representation of the occurrence of motifs, notably by projecting features representing forward and reverse-complement motifs to similar positions in the lower dimensional feature space output by the layer.

bioinformatics↗