Biology subjects

Michael, R.

Publications and source records attributed to Michael, R..

6 recordsLinked to original sources

bioRxiv2026

APOSM: Pairwise preference learning improves generative small-molecule design

Small-molecule lead refinement is constrained by the cost of synthesizing and assaying candidates, making the surrogate models that prioritize compounds for experimental testing central to the design process. The reliability of such surrogates is limited by the noise and sparsity of screening measurements. We show that training the surrogate on pairwise comparisons between candidate molecules, rather than on absolute predicted scores, yields a substantially more reliable signal for active candidate selection in this regime. We develop APOSM, an active-learning algorithm that combines a fragment-based generator, a pairwise message-passing graph neural network surrogate, and probabilistic ranking inside a batched acquisition loop. On the Practical Molecular Optimization benchmark and a GPCR ligand rediscovery task, APOSM improves target attainment and sampling efficiency over unguided fragment-based optimization, the Graph-GA genetic algorithm, and a pointwise-regression ablation, with the largest gains on tasks where absolute scores are hardest to calibrate.

bioinformatics↗

bioRxiv2024

Label-Free Visualization and Segmentation of Endothelial Cell Mitochondria Using Holotomographic Microscopy and U-Net

Understanding the physiological processes underlying age-related cardiovascular disease (CVD) requires examination of endothelial cell (EC) mitochondrial networks, because mitochondrial function and adenosine triphosphate production are crucial in EC metabolism, and consequently influence CVD progression. Although current biochemical assays and immunofluorescence microscopy can reveal how mitochondrial function influences cellular metabolism, they cannot achieve live observation and tracking changes in mitochondrial networks through fusion and fission events. Holotomographic microscopy (HTM) has emerged as a promising technique for real-time, label-free visualization of ECs and their organelles, such as mitochondria. This non-destructive, non-interfering live cell imaging method offers unprecedented opportunities to observe mitochondrial network dynamics. However, because existing image processing tools based on immunofluorescence microscopy techniques are incompatible with HTM images, a machine-learning model is required. Here, we developed a model using a U-net learner with a Resnet18 encoder to identify four classes within HTM images: mitochondrial networks, cell borders, ECs, and background. This method accurately identifies mitochondrial structures and positions. With high accuracy and similarity metrics, the output image successfully provides visualization of mitochondrial networks within HTM images of ECs. This approach enables the study of mitochondrial networks and their effects, and holds promise in advancing understanding of CVD mechanisms.

bioengineering↗

bioRxiv2023

FLOP: Tasks for Fitness Landscapes Of Protein wildtypes

Protein engineering has the potential to create optimized protein variants with improved properties and function. An initial step in the protein optimization process typically consists of a search among natural (wildtype) sequences to find the naturally occurring proteins with the most desirable properties. Promising candidates from this initial discovery phase then form the basis of the second step: a more local optimization procedure, exploring the space of variants separated from this candidate by a number of mutations. While considerable progress has been made on evaluating machine learning methods on single protein datasets, benchmarks of data-driven approaches for global fitness landscape exploration are still lacking. In this paper, we have carefully curated a representative benchmark dataset, which reflects industrially relevant scenarios for the initial wildtype discovery phase of protein engineering. We focus on exploration within a protein family, and investigate the downstream predictive power of various protein representation paradigms, i.e., protein language model-based representations, structure-based representations, and evolution-based representations. Our benchmark highlights the importance of coherent split strategies, and how we can be misled into overly optimistic estimates of the state of the field. The codebase and data can be accessed via https://github.com/petergroth/FLOP.

bioinformatics↗

bioRxiv2023

Assessing the performance of protein regression models

To optimize proteins for particular traits holds great promise for industrial and pharmaceutical purposes. Machine Learning is increasingly applied in this field to predict properties of proteins, thereby guiding the experimental optimization process. A natural question is: How much progress are we making with such predictions, and how important is the choice of regressor and representation? In this paper, we demonstrate that different assessment criteria for regressor performance can lead to dramatically different conclusions, depending on the choice of metric, and how one defines generalization. We highlight the fundamental issues of sample bias in typical regression scenarios and how this can lead to misleading conclusions about regressor performance. Finally, we make the case for the importance of calibrated uncertainty in this domain.

bioinformatics↗

bioRxiv2023

A generalizable Cas9/sgRNA prediction model using machine transfer learning with small high-quality datasets

The CRISPR/Cas9 nuclease from Streptococcus pyogenes (SpCas9) can be used with single guide RNAs (sgRNAs) as a sequence-specific antimicrobial agent and as a genome-engineering tool. However, current bacterial sgRNA activity models poorly predict SpCas9/sgRNA activity and are not generalizable, possibly because the underlying datasets used to train the models do not accurately measure SpCas9/sgRNA cleavage activity and cannot distinguish cleavage activity from toxicity. We solved this problem by using a two-plasmid positive selection system to generate high-quality biologically-relevant data that more accurately reports on SpCas9/sgRNA cleavage activity and that separates activity from toxicity. We developed a new machine transfer learning architecture (crisprHAL) that can be trained on existing datasets and that shows marked improvements in sgRNA activity prediction accuracy when transfer learning is used with small amounts of high-quality data. The crisprHAL model recapitulates known SpCas9/sgRNA-target DNA interactions and provides a pathway to a generalizable sgRNA bacterial activity prediction tool.

bioinformatics↗

bioRxiv2019

The Nonlinear Relations that Predict Influenza Viral Dynamics, CD8+ T cell-Mediated Clearance, Lung Pathology, and Disease Severity

Influenza viruses cause a significant amount of morbidity and mortality. Understanding host immune control efficacy and how different factors influence lung injury and disease severity are critical. Here, we established dynamical connections between viral loads, infected cells, CD8+ T cell-mediated clearance, lung injury, inflammation, and disease severity using an integrative model-experiment exchange. The model was validated through CD8 depletion and whole lung histomorphometry, which showed that the infected area matched the model-predicted infected cell dynamics and that the resolved area paralleled the relative CD8 dynamics. Inflammation could further be predicted by the infected cell dynamics, and additional analyses revealed nonlinear relations between lung injury, inflammation, and disease severity. These links between important pathogen kinetics and host pathology enhance our ability to forecast disease progression, potential complications, and therapeutic efficacy.

microbiology↗