Price, G. | bioRxiv Science

bioRxiv2021

LM-GVP: A Generalizable Deep Learning Framework for Protein Property Prediction from Sequence and Structure

Proteins perform many essential functions in biological systems and can be successfully developed as bio-therapeutics. It is invaluable to be able to predict their properties based on a proposed sequence and structure. In this study, we developed a novel generalizable deep learning framework, LM-GVP, composed of a protein Language Model (LM) and Graph Neural Network (GNN) to leverage information from both 1D amino acid sequences and 3D structures of proteins. Our approach outperformed the state-of-the-art protein LMs on a variety of property prediction tasks including fluorescence, protease stability, and protein functions from Gene Ontology (GO). We also illustrated insights into how a GNN prediction head can guide the protein LM to better leverage structural information. We envision that our deep learning framework will be generalizable to many protein property prediction problems to greatly accelerate protein engineering and drug development.

synthetic biology↗

bioRxiv2021

Generation, quality control, and analysis of the first genomically humanised knock-in mice for the ALS/FTD genes SOD1, TARDBP (TDP-43), and FUS

Amyotrophic lateral sclerosis - frontotemporal dementia spectrum disorder (ALS/FTD) is a complex neurodegenerative disease; up to 10% of cases are familial, usually arising from single dominant mutations in >30 causative genes. Transgenic mouse models that overexpress human ALS/FTD causative genes have been the preferred organism for in vivo modelling. However, while conferring human protein biochemistry, these overexpression models are not ideal for dosage-sensitive proteins such as TDP-43 or FUS. We have created three next-generation genomically humanised knock-in mouse models for ALS/FTD research, by replacing the entire mouse coding region of Sod1, Tardbp (TDP-43) and Fus, with their human orthologues to preserve human protein biochemistry, with exons and introns intact to enable future modelling of coding or non-coding mutations and variants and to preserve human splice variants. In generating these mice, we have established a new-standard of quality control: we demonstrate the utility of indirect capture for enrichment of a region of interest followed by Oxford Nanopore sequencing for robustly characterising large knock-in alleles. This approach confirmed that targeting occurred at the correct locus and to map homologous recombination events. Furthermore, extensive expression data from the three lines shows that homozygous humanised animals only express human protein, at endogenous levels. Characterisation of humanised FUS animals showed that they are phenotypically normal compared to wildtype littermates throughout their lifespan. These humanised mouse strains are critically needed for preclinical assessment of interventions, such as antisense oligonucleotides (ASOs), to modulate expression levels in patients, and will serve as templates for the addition of human ALS/FTD mutations to dissect disease pathomechanisms.

neuroscience↗