Biology subjects

Copoiu, L.

Publications and source records attributed to Copoiu, L..

8 recordsLinked to original sources

bioRxiv2025

AbBFN2: A flexible antibody foundation model based on Bayesian Flow Networks

Antibody engineering is marked by diverse data and desiderata, making it a prime candidate for multi-objective design, but is commonly tackled as a series of sequential optimisation tasks. Here, we present AbBFN2, a generative antibody foundation model trained on paired antibody sequences as well as genetic and biophysical metadata. This is achieved using the Bayesian Flow Network paradigm, which allows unified modelling of diverse data sources and flexible conditional generation at inference time. By virtue of its rich set of features and architectural flexibility, AbBFN2 can be adapted to a number of tasks commonly tackled by individual models, consolidating traditional computational pipelines into a single step. We demonstrate the adaptability of AbBFN2 using sequence inpainting, humanisation, biophysical property optimisation, and conditional de novo library generation of antibodies with rare attributes as example tasks. By removing the need for task-specific training, we hope that AbBFN2 will accelerate machine learning-based antibody design and development workflows.

bioinformatics↗

bioRxiv2024

Protein Sequence Modelling with Bayesian Flow Networks

Exploring the vast and largely uncharted territory of amino acid sequences is crucial for understanding complex protein functions and the engineering of novel therapeutic proteins. Whilst generative machine learning has advanced protein sequence modelling, no existing approach is proficient for both unconditional and conditional generation. In this work, we propose that Bayesian Flow Networks (BFNs), a recently introduced framework for generative modelling, can address these challenges. We present ProtBFN, a 650M parameter model trained on protein sequences curated from UniProtKB, which generates natural-like, diverse, structurally coherent, and novel protein sequences, significantly outperforming leading autoregressive and discrete diffusion models. Further, we fine-tune ProtBFN on heavy chains from the Observed Antibody Space (OAS) to obtain an antibody-specific model, AbBFN, which we use to evaluate zero-shot conditional generation capabilities. AbBFN is found to be competitive with, or better than, antibody-specific BERT-style models, when applied to predicting individual framework or complimentary determining regions (CDR).

bioinformatics↗

bioRxiv2024

Are Genomic Language Models All You Need? Exploring Genomic Language Models on Protein Downstream Tasks

Large language models, trained on enormous corpora of biological sequences, are state-of-the-art for downstream genomic and proteomic tasks. Since the genome contains the information to encode all proteins, genomic language models (gLMs) hold the potential to make downstream predictions not only about DNA sequences, but also about proteins. However, the performance of gLMs on protein tasks remains unknown, due to few tasks pairing proteins with the coding DNA sequences (CDS) that can be processed by gLMs. In this work, we curated five such datasets and used them to evaluate the performance of gLMs and proteomic language models (pLMs). We show that gLMs are competitive and even outperform their pLMs counterparts on some tasks. The best performance was achieved using the retrieved CDS compared to sampling strategies. We found that training a joint genomic-proteomic model outperforms each individual approach, showing that they capture different but complementary sequence representations, as we demonstrate through model interpretation of their embeddings. Lastly, we explored different genomic tokenization schemes to improve downstream protein performance. We trained a new Nucleotide Transformer (50M) foundation model with 3mer tokenization that outperforms its 6mer counterpart on protein tasks while maintaining performance on genomics tasks. The application of gLMs to proteomics offers the potential to leverage rich CDS data, and in the spirit of the central dogma, the possibility of a unified and synergistic approach to genomics and proteomics. We make our inference code, model weights and datasets available.

bioinformatics↗

bioRxiv2023

Contrasting Sequence with Structure: Pre-training Graph Representations with PLMs

Understanding protein function is vital for drug discovery, disease diagnosis, and protein engineering. While Protein Language Models (PLMs) pre-trained on vast protein sequence datasets have achieved remarkable success, equivalent Protein Structure Models (PSMs) remain underrepresented. We attribute this to the relative lack of high-confidence structural data and suitable pre-training objectives. In this context, we introduce BioCLIP, a contrastive learning framework that pre-trains PSMs by leveraging PLMs, generating meaningful per-residue and per-chain structural representations. When evaluated on tasks such as protein-protein interaction, Gene Ontology annotation, and Enzyme Commission number prediction, BioCLIP-trained PSMs consistently outperform models trained from scratch and further enhance performance when merged with sequence embeddings. Notably, BioCLIP approaches, or exceeds, specialized methods across all benchmarks using its singular pre-trained design. Our work addresses the challenges of obtaining quality structural data and designing self-supervised objectives, setting the stage for more comprehensive models of protein function. Source code is publicly available2.

bioinformatics↗

bioRxiv2023

LightMHC: A Light Model for pMHC Structure Prediction with Graph Neural Networks

The peptide-major histocompatibility complex (pMHC) is a crucial protein in cell-mediated immune recognition and response. Accurate structure prediction is potentially beneficial for protein interaction prediction and therefore helps immunotherapy design. However, predicting these structures is challenging due to the sequential and structural variability. In addition, existing pre-trained models such as AlphaFold 2 require expensive computation thus inhibiting high throughput in silico peptide screening. In this study, we propose LightMHC: a lightweight model (2.2M parameters) equipped with attention mechanisms, graph neural networks, and convolutional neural networks. LightMHC predicts full-atom pMHC structures from amino-acid sequences alone, without template structures. The model achieved comparable or superior performance to AlphaFold 2 and ESMFold (93M and 15B parameters respectively), with five-fold acceleration (6.65 seconds/sample for LightMHC versus 36.82 seconds/sample for AlphaFold 2), potentially offering a valuable tool for immune protein structure prediction and immunotherapy design.

bioinformatics↗

bioRxiv2022

Peptide-MHC Structure Prediction With Mixed Residue and Atom Graph Neural Network

Neoantigen-targeting vaccines have achieved breakthrough success in cancer immunotherapy by eliciting immune responses against neoantigens, which are proteins uniquely produced by cancer cells. During the immune response, the interactions between peptides and major histocompatibility complexes (MHC) play an important role as peptides must be bound and presented by MHC to be recognised by the immune system. However, only limited experimentally determined peptide-MHC (pMHC) structures are available, and in-silico structure modelling is therefore used for studying their interactions. Current approaches mainly use Monte Carlo sampling and energy minimisation, and are often computationally expensive. On the other hand, the advent of large high-quality proteomic data sets has led to an unprecedented opportunity for deep learning-based methods with pMHC structure prediction becoming feasible with these trained protein folding models. In this work, we present a graph neural network-based model for pMHC structure prediction, which takes an amino acid-level pMHC graph and an atomic-level peptide graph as inputs and predicts the peptide backbone conformation. With a novel weighted reconstruction loss, the trained model achieved a similar accuracy to AlphaFold 2, requiring only 1.7M learnable parameters compared to 93M, representing a more than 98% reduction in the number of required parameters.

bioinformatics↗

bioRxiv2021

Predicted structural mimicry of spike receptor-binding motifs from highly pathogenic human coronaviruses

Viruses often encode proteins that mimic host proteins in order to facilitate infection. Little work has been done to understand the potential mimicry of the SARS-CoV-2, SARS-CoV, and MERS-CoV spike proteins, particularly the receptor-binding motifs, which could be important in determining tropism of the virus. Here, we use structural bioinformatics software to characterize potential mimicry of the three coronavirus spike protein receptor-binding motifs. We utilize sequence-independent alignment tools to compare structurally known or predicted three-dimensional protein models with the receptor-binding motifs and verify potential mimicry with protein docking simulations. Both human and non-human proteins were found to be similar to all three receptor-binding motifs. Similarity to human proteins may reveal which pathways the spike protein is co-opting, while analogous non-human proteins may indicate shared host interaction partners and overlapping antibody cross-reactivity. These findings can help guide experimental efforts to further understand potential interactions between human and coronavirus proteins. HighlightsO_LIPotential coronavirus spike protein mimicry revealed by structural comparison C_LIO_LIHuman and non-human protein potential interactions with virus identified C_LIO_LIPredicted structural mimicry corroborated by protein-protein docking C_LIO_LIEpitope-based alignments may help guide vaccine efforts C_LI Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=111 SRC="FIGDIR/small/441187v1_ufig1.gif" ALT="Figure 1"> View larger version (22K): org.highwire.dtl.DTLVardef@1f09454org.highwire.dtl.DTLVardef@19a5557org.highwire.dtl.DTLVardef@158d3fdorg.highwire.dtl.DTLVardef@c59511_HPS_FORMAT_FIGEXP M_FIG C_FIG

bioinformatics↗

bioRxiv2019

The midbody interactome reveals new unexpected roles for PP1 phosphatases in cytokinesis

The midbody is an organelle assembled at the intercellular bridge that connects the two daughter cells at the end of mitosis. It is composed of a multitude of proteins, organized in a precise and stereotyped pattern, that control the final separation of the daughter cells and prevent incorrect genome segregation. Furthermore, recent evidence indicates that the midbody is involved in many other important processes, including cell fate, pluripotency, apical-basal polarity, tissue organization, and cilium and lumen formation. Understanding the regulation and interactions of midbody proteins is therefore essential to unravel how this organelle executes its multiple functions. Here, we report the first experimentally-based characterization of the intricate midbody protein-protein interaction network (interactome), which identifies a plethora of novel interactions and provides an extremely valuable resource for dissecting the multiple roles of the midbody. Initial analysis of this interactome already revealed that PP1{beta}-MYPT1 phosphatase regulates microtubule dynamics in late cytokinesis and de-phosphorylates the kinesin component MKLP1/KIF23 of the centralspindlin complex, a key cytokinesis regulator. This de-phosphorylation antagonizes Aurora B kinase in order to modify the functions of centralspindlin and its interactions in late cytokinesis. Our findings expand the repertoire of PP1 functions during mitosis and indicate that spatiotemporal changes in the distribution of kinases and counteracting phosphatases finely tune the activity of cytokinesis proteins.

cell biology↗