Biology subjects

Gligorijevic, V.

Publications and source records attributed to Gligorijevic, V..

At least 19 recordsLinked to original sources

bioRxiv2026

MIMOSA: Guiding De Novo Antibody Design with Natural Interaction Fingerprints

De novo antibody design promises to transform therapeutic antibody and nanobody discovery. Yet, success remains inconsistent across targets, with generative models often yielding no experimentally validated binders. Here we introduce MIMOSA (MIMic-Oriented Structural Antibody generation), a model-agnostic, inference-time framework that constrains diffusion-based antibody design pipelines to reproduce the interaction geometry and chemistry of a target's known cognate binder interface. Rather than searching for a productive interface from scratch, MIMOSA constrains generation around residue identities and geometries already known to support binding, while allowing the generative model to complete the surrounding antibody interface. We test MIMOSA on different interface topologies: contiguous motifs that fit within a single CDR loop and spatially dispersed hotspots distributed across the paratope. Applied without retraining to two architecturally distinct generators (RFAntibody and BoltzGen), this framework delivers experimentally validated binders-achieving sub-micromolar to single-digit nanomolar affinities-at 20-26% hit rates on three targets on which unconstrained methods produce zero or near-zero binders: KEAP1, uPA, and IL-8. By turning any target with a structurally characterised cognate binder into an accessible de novo design problem, this framework broadens the practical reach of generative antibody design to targets where unconstrained approaches currently fail.

bioinformatics↗

bioRxiv2025

CDR Conformation Aware Antibody Sequence Design with ConformAb

Antibody lead optimization methods seek to enhance a lead candidates therapeutic properties through targeted sequence mutation. However, the mutations introduced during this process can inadvertently induce structural changes that disrupt binding, particularly by altering CDR loop conformations which destabilize CDR-target binding interactions. To address this, we present ConformAb, a guided discrete diffusion model for designing antibody sequences that explicitly conform to the CDR canonical conformation of the lead. With seed sequences as the starting point, we show that our method is capable of generating 3- 5x better binders that conform to the same CDR backbone structure in a one-shot design setup, and in some cases, binders with better affinity than seed repertoire picks without any target-specific data included in training. Across the targets tested, ConformAb demonstrated binding rates ranging from 15 -60% in wet lab experiments, a result obtained using fewer than 100 designs for each target. ConformAb offers a unique one-shot approach for antibody lead optimization in data-scarce scenarios where experimental/repertoire data cannot be leveraged for model training.

biophysics↗

bioRxiv2025

Property Enhancer - a data efficient multi-objective approach for functional antibody optimization

In-silico antibody lead optimization remains challenging due to scarce high-quality data, costly experimental validation, and the need to jointly optimize multiple developability properties. Discovery workflows often rely on high-throughput phage, ribosome or yeast display experiments, which yield large but noisy datasets; as leads emerge, strategies shift to low-throughput assays which are precise, yet unscalable. Deep-learning and language-model approaches are hindered by such limited, unreliable measurements. We introduce Property Enhancer (PropEn), a data-efficient framework for low-data, heterogeneous regimes that can simultaneously optimize multiple antibody properties. PropEn proposes a matching-based augmentation that expands the training data with sequence pairs differing by only a few mutations; within each pair the second sequence improves the target value, providing an implicit optimization signal. Extensive in silico and in vitro tests show 10-39x affinity gains across four targets and nine leads, and enable joint multi-property optimization, positioning PropEn as a scalable, general solution.

molecular biology↗

bioRxiv2025

DyAb: sequence-based antibody design and property prediction in a low-data regime

Protein therapeutic design and property prediction are frequently hampered by data scarcity. Here we propose a new model, DyAb, that addresses these issues by leveraging a pair-wise representation to predict differences in protein properties, rather than absolute values. DyAb is built on top of a pre-trained protein language model and achieves a Spearman rank correlation of up to 0.85 on binding affinity prediction across molecules targeting three different antigens (EGFR, IL-6, and an internal target), given as few as 100 training data. We employ DyAb in two design contexts: as a ranking model to score combinations of known mutations, and combined with a genetic algorithm to generate new sequences. Our method consistently generates novel antibody candidates with high binding rates, including designs that improve on the binding affinity of the lead molecule by more than ten-fold. DyAb represents a powerful tool for engineering therapeutic protein properties in low data regimes common in early-stage drug development.

bioengineering↗

bioRxiv2025

deepNGS Navigator: Exploring antibody NGS datasets using deep contrastive learning

High-throughput sequencing uncovers how B-cells adapt and evolve in response to antigens by generating B-cell receptor (BCR) sequences at an unprecedented scale. As BCR datasets grow to be millions of sequences, using efficient computational methods becomes crucial for analyzing and understanding complex patterns within the data. One important aspect of antibody sequence analysis is detecting clonal families or clusters of related sequences, whether they come from immunization, synthetic libraries or even ML-generated datasets. Such analysis helps us to understand how sequences are evolutionarily connected, and how they might have been selected or evolved. Here we introduce deepNGS Navigator, a computational tool that leverages language models and contrastive learning to transform large datasets of antibody sequences into 2D representations. The resulting 2D maps offer an intuitive visualization of overall diversity of input datasets, which can be clustered based on the sequence distances and their densities across the map. Beyond grouping related sequences, the 2D maps can also point to evolutionary trajectories and capture mutational patterns among closely related sequences. By analyzing properties like charge, hydrophobicity, number of sequence neighbors, and read counts, the maps highlight which clusters are most promising for further investigation while also detecting anomalies or noisy sequences with higher risk. We demonstrate deepNGS Navigators utilities on various datasets, including: 1) a synthetic library from a yeast display targeting HER2, 2) a machine learning-generated dataset with a hierarchical tree structure, 3) NGS sequences from a llama immunized against COVID RBD, 4) human naive and memory B-cell sequences, and 5) an insilico dataset simulating B-cell clonal lineages.

bioinformatics↗

bioRxiv2024

Generating All-Atom Protein Structure from Sequence-Only Training Data

AbstractO_ST_ABSPurposeC_ST_ABSDesigning proteins with atomic-level functional control remains a central challenge in de novo design, exacerbated by limitations in structural data availability. MethodsWe introduce PLAID (Protein Latent Induced Diffusion), a generative model that efficiently co-generates discrete sequence and all-atom structure by sampling directly in the shared sequence-structure latent of a pretrained sequence-to-structure predictor. Unlike existing de novo generative models, PLAID trains its diffusion model exclusively on sequences, expanding effective training corpora by 2-4 orders of magnitude relative to structural databases. Using classifier-free guidance, PLAID supports controllable generation based on function (Gene Ontology) and organism keywords. ResultsIn silico, PLAID can unconditionally generate all-atom structures without explicit structural supervision during diffusion model training. Function conditioned proteins can recapitulates catalytic side-chain positions for residues at non-adjacent positions, and transmembrane proteins with expected hydrophobicity patterns and predicted topologies. We further experimentally validate that PLAID can be prompted to generate heme binding proteins with high sequence novelty. ConclusionOverall, PLAID unifies sequence-scale training with atomic-level generation, enabling more precise functional control in protein design. Code and weights: github.com/amyxlu/plaid.

bioinformatics↗

bioRxiv2024

Tokenized and Continuous Embedding Compressions of Protein Sequence and Structure

Existing protein machine learning representations typically model either the sequence or structure distribution, with the other modality implicit. The latent space of sequence-to-structure prediction models such as ESMFold represents the joint distribution of sequence and structure; however, we find these embeddings to exhibit massive activations, whereby some channels have values 3000x higher than others, regardless of the input. Further, on continuous compression schemes, ESMFold embeddings can be reduced by a factor of 128x along the channel and 8x along the length, while retaining structure information at <2[A] scale accuracy, and performing competitively on protein function and localization benchmarks. On discrete compression schemes, we construct a tokenized all-atom structure vocabulary that retains high reconstruction accuracy, thus introducing a tokenized representation of all-atom structure that can be obtained from sequence alone. We term this series of embeddings as CHEAP (Compressed Hourglass Embedding Adaptations of Proteins) embeddings, obtained via the HPCT (Hourglass Protein Compression Transformer) architecture. CHEAP is a compact representation of both protein structure and sequence, sheds light on information content asymmetries between sequence and structure, democratizes representations captured by large models, and is designed to have flexible downstream applications such as generation, search, and prediction.

bioinformatics↗

bioRxiv2024

Cramming Protein Language Model Training in 24 GPU Hours

Protein language models (pLMs) are ubiquitous across biological machine learning research, but state-of-the-art models like ESM2 take hundreds of thousands of GPU hours to pre-train on the vast protein universe. Resource requirements for scaling up pLMs prevent fundamental investigations into how optimal modeling choices might differ from those used in natural language. Here, we define a "cramming" challenge for pLMs and train performant models in 24 hours on a single GPU. By re-examining many aspects of pLM training, we are able to train a 67 million parameter model in a single day that achieves comparable performance on downstream protein fitness landscape inference tasks to ESM-3B, a model trained for over 15, 000x more GPU hours than ours. We open source our library1 for training and inference, LBSTER: Language models for Biological Sequence Transformation and Evolutionary Representation.

bioengineering↗

bioRxiv2023

Mixed structure- and sequence-based approach for protein graph neural networks with application to antibody developability prediction

There are hundreds of thousands of known proteins but significantly fewer unique protein folds. Furthermore, proteins often have conserved and even repeating geometric patterns, which can be captured by models of protein structure and function. In this work, we use Delaunay tessellations and -shapes, which capture these conserved geometric patterns, to define graph adjacency in Graph Convolutional Networks (GCN). We demonstrate the utility of the resulting GCN model on antibody developability prediction. Compared to the frequently used graph adjacencies based on k-nearest-neighbors or a fixed cutoff radius, the Delaunay tessellation and -shapes better capture residue-specific interactions at a lower computational cost for a given system size. The resulting models achieve state-of-the-art performance on an antibody developability prediction task. Finally, we propose an extension of the model which does not require known or predicted structures but uses an "archetypical" antibody structure to infer likely contacts.

biophysics↗

bioRxiv2022

Automated Protein Function Description for NovelClass Discovery

Knowledge of protein function is necessary for understanding biological systems, but the discovery of new sequences from high-throughput sequencing technologies far outpaces their functional characterization. Beyond the problem of assigning newly sequenced proteins to known functions, a more challenging issue is discovering novel protein functions. The space of possible functions becomes unlimited when considering designed proteins. Protein function prediction, as it is framed in the case of Gene Ontology term prediction, is a multilabel classification problem with a hierarchical label space. However, this framing does not provide guiding principles for discovering completely novel functions. Here we propose a neural machine translation model in order to generate descriptions of protein functions in natural language. In this way, instead of making predictions in a limited label space, our model generates descriptions in the language space, and thus is capable of composing novel functions. Given the novelty of our approach, we design metrics to evaluate the performance of our model: correctness, specificity and robustness. We provide results of our model in the zero-shot classification setting, scoring functional descriptions that the model has not seen before for proteins that have limited homology to those in the training set. Finally, we show generated function descriptions compared to ground truth descriptions for qualitative evaluation.

genomics↗

bioRxiv2022

EquiFold: Protein Structure Prediction with a Novel Coarse-Grained Structure Representation

Designing proteins to achieve specific functions often requires in silico modeling of their properties at high throughput scale and can significantly benefit from fast and accurate protein structure prediction. We introduce EquiFold, a new end-to-end differentiable, SE(3)-equivariant, all-atom protein structure prediction model. EquiFold uses a novel coarse-grained representation of protein structures that does not require multiple sequence alignments or protein language model embeddings, inputs that are commonly used in other state-of-the-art structure prediction models. Our method relies on geometrical structure representation and is substantially smaller than prior state-of-the-art models. In preliminary studies, EquiFold achieved comparable accuracy to AlphaFold but was orders of magnitude faster. The combination of high speed and accuracy make EquiFold suitable for a number of downstream tasks, including protein property prediction and design.

bioinformatics↗

bioRxiv2022

TM-Vec: template modeling vectors for fast homology detection and alignment

Exploiting sequence-structure-function relationships in molecular biology and computational modeling relies on detecting proteins with high sequence similarities. However, the most commonly used sequence alignment-based methods, such as BLAST, frequently fail on proteins with low sequence similarity to previously annotated proteins. We developed a deep learning method, TM-Vec, that uses sequence alignments to learn structural features that can then be used to search for structure-structure similarities in large sequence databases. We train TM-Vec to accurately predict TM-scores as a metric of structural similarity for pairs of structures directly from sequence pairs without the need for intermediate computation or solution of structures. For remote homologs (sequence similarity [≤] 10%) that are highly structurally similar (TM-score ? 0.6), we predict TM-scores within 0.026 of their value computed by TM-align. TM-Vec outperforms traditional sequence alignment methods and performs similar to structure-based alignment methods. TM-Vec was trained on the CATH and SwissModel structural databases and it has been tested on carefully curated structure-structure alignment databases that were designed specifically to test very remote homology detection methods. It scales sub-linearly for search against large protein databases and is well suited for discovering remotely homologous proteins.

bioinformatics↗

bioRxiv2022

Comprehensive function annotation of metagenomes and microbial genomes using a deep learning-based method

Comprehensive protein function annotation is essential for understanding microbiome-related disease mechanisms in the host organisms. Still, a large portion of human gut microbial proteins lack functional annotation. Here, we have developed a new metagenome analysis workflow integrating de novo genome reconstruction, taxonomic profiling and deep learning-based functional annotations from DeepFRI. We validate DeepFRI functional annotations by comparing them to orthology-based annotations from eggNOG on a set of 1,070 infant metagenome samples from the DIABIMMUNE cohort. Using the workflow, we have generated a sequence catalogue of 1.9 million non-redundant microbial genes. The functional annotations revealed 70% concordance between GO annotations predicted by DeepFRI and eggNOG. However, DeepFRI improved the annotation coverage, with 99% of the gene catalogue obtaining GO molecular function annotations, albeit less specific compared to eggNOG. Additionally, we construct pan-genomes in a reference-free manner using high-quality metagenome assembled genomes (MAGs) and analyse the associated annotations. eggNOG annotated more genes on well-studied organisms such as Escherichia coli while DeepFRI was less sensitive to taxa. This workflow will contribute to novel understanding of the functional signature of the human gut microbiome in health and disease as well as guide future metagenomics studies.

bioinformatics↗

bioRxiv2022

Predicting recognition between T cell receptors and epitopes using contextualized motifs

We introduce TCRconv, a deep learning model for predicting recognition between T-cell receptors and epitopes. TCRconv uses a deep protein language model and convolutions to extract contextualized motifs and provides state-of-the-art TCR-epitope prediction accuracy. Using TCR repertoires from COVID-19 patients, we demonstrate that TCRconv can provide insight into T-cell dynamics and phenotypes during the disease.

bioinformatics↗

bioRxiv2022

Sequence-structure-function relationships in the microbial protein universe

Abstract / SummaryFor the past half-century, structural biologists relied on the notion that similar protein sequences give rise to similar structures and functions. While this assumption has driven research to explore certain parts of the protein universe, it disregards spaces that dont rely on this assumption. Here we explore areas of the protein universe where similar protein functions can be achieved by different sequences and different structures. We predict [~]200,000 structures for diverse protein sequences from 1,003 representative genomes1 across the microbial tree of life, and annotate them functionally on a per-residue basis. Structure prediction is accomplished using the World Community Grid, a large-scale citizen science initiative. The resulting database of structural models is complementary to the AlphaFold database, with regards to domains of life as well as sequence diversity and sequence length. We identify 148 novel folds and describe examples where we map specific functions to structural motifs. We also show that the structural space is continuous and largely saturated, highlighting the need for shifting the focus from obtaining structures to putting them into context, to transform all branches of biology, including a shift from sequence-based to sequence-structure-function based meta-omics analyses.

bioinformatics↗

bioRxiv2021

Function-guided protein design by deep manifold sampling

Protein design is challenging because it requires searching through a vast combinatorial space that is only sparsely functional. Self-supervised learning approaches offer the potential to navigate through this space more effectively and thereby accelerate protein engineering. We introduce a sequence denoising autoencoder (DAE) that learns the manifold of protein sequences from a large amount of potentially unlabelled proteins. This DAE is combined with a function predictor that guides sampling towards sequences with higher levels of desired functions. We train the sequence DAE on more than 20M unlabeled protein sequences spanning many evolutionarily diverse protein families and train the function predictor on approximately 0.5M sequences with known function labels. At test time, we sample from the model by iteratively denoising a sequence while exploiting the gradients from the function predictor. We present a few preliminary case studies of protein design that demonstrate the effectiveness of this proposed approach, which we refer to as "deep manifold sampling", including metal binding site addition, function-preserving diversification, and global fold change.

bioinformatics↗

bioRxiv2020

Protein Structural Alignments From Sequence

Computing sequence similarity is a fundamental task in biology, with alignment forming the basis for the annotation of genes and genomes and providing the core data structures for evolutionary analysis. Standard approaches are a mainstay of modern molecular biology and rely on variations of edit distance to obtain explicit alignments between pairs of biological sequences. However, sequence alignment algorithms struggle with remote homology tasks and cannot identify similarities between many pairs of proteins with similar structures and likely homology. Recent work suggests that using machine learning language models can improve remote homology detection. To this end, we introduce DeepBLAST, that obtains explicit alignments from residue embeddings learned from a protein language model integrated into an end-to-end differentiable alignment framework. This approach can be accelerated on the GPU architectures and outperforms conventional sequence alignment techniques in terms of both speed and accuracy when identifying structurally similar proteins.

bioinformatics↗

bioRxiv2020

NetQuilt: Deep Multispecies Network-based Protein Function Prediction using Homology-informed Network Similarity

Transferring knowledge between species is challenging: different species contain distinct proteomes and cellular architectures, which cause their proteins to carry out different functions via different interaction networks. Many approaches to proteome and biological network functional annotation use sequence similarity to transfer knowledge between species. These similarity-based approaches cannot produce accurate predictions for proteins without homologues of known function, as many functions require cellular or organismal context for meaningful function prediction. In order to supply this context, network-based methods use protein-protein interaction (PPI) networks as a source of information for inferring protein function and have demonstrated promising results in function prediction. However, the majority of these methods are tied to a network for a single species, and many species lack biological networks. In this work, we integrate sequence and network information across multiple species by applying an IsoRank-derived network alignment algorithm to create a meta-network profile of the proteins of multiple species. We then use this integrated multispecies meta-network as input features to train a maxout neural network with Gene Ontology terms as target labels. Our multispecies approach takes advantage of more training examples, and more diverse examples from multiple organisms, and consequently leads to significant improvements in function prediction performance. Further, we evaluate our approach in a setting in which an organisms PPI network is left out, using other organisms network information and sequence homology in order to make predictions for the left-out organism, to simulate cases in which a newly sequenced species has no network information available.

bioinformatics↗