Aryan Bhasin

Aryan Bhasin

Auto-generated profile, based on publicly available information

Aryan Bhasin

InstaDeep Ltd

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
17 July 2025 in bioRxiv
Verified
Preprint
Open access
ABAryan Bhasin
Aryan Bhasin
FSFrancesco Saccon
Francesco Saccon
CCCallum Canavan
Callum Canavan
ARAndrew Robson
Andrew Robson
JEJoao Euko
Joao Euko
YFYunguan Fu
Yunguan Fu
Since the emergence of SARS-CoV-2, numerous studies have investigated antibody interactions with viral variants in vitro , and several datasets have been curated to compile available protein structures and experimental measurements. However, existing data remain fragmented, limiting their utility for the development and validation of machine learning models for antibody–antigen interaction prediction. Here, we present CoV-UniBind, a unified database comprising over 75,000 entries of SARS-CoV-2 antibody–antigen sequence, binding, and structural data, integrated and standardised from three public sources and multiple peer-reviewed publications. To demonstrate its utility, we benchmarked multiple protein folding and inverse folding models across tasks relevant to antibody design and vaccine development. We expect CoV-UniBind to facilitate future computational efforts in antibody and vaccine development against SARS-CoV-2. Availability and implementation: The curated datasets, structures, model scores and antibody synonyms are free to download at https://huggingface.co/datasets/InstaDeepAI/cov-unibind . Folded structures are available upon request.

2 January 2026 in Bioinformatics Advances
Verified
Journal Article
Open access
ABAryan Bhasin
Aryan Bhasin
FSFrancesco Saccon
Francesco Saccon
CCCallum Canavan
Callum Canavan
ARAndrew Robson
Andrew Robson
JEJoao Euko
Joao Euko
YFYunguan Fu
Yunguan Fu
Summary: Since the emergence of SARS-CoV-2, numerous studies have investigated antibody interactions with viral variants in vitro, and several datasets have been curated to compile available protein structures and experimental measurements. However, existing data remain fragmented, limiting their utility for the development and validation of machine learning models for antibody–antigen interaction prediction. Here, we present CoV-UniBind, a unified database comprising over 75 000 entries of SARS-CoV-2 antibody–antigen sequence, binding, and structural data, integrated and standardized from three public sources and multiple peer-reviewed publications. To demonstrate its utility, we benchmarked multiple protein folding, inverse folding, and language models across tasks relevant to antibody design and vaccine development. We expect CoV-UniBind to facilitate future computational efforts in antibody and vaccine development against SARS-CoV-2. Availability and implementation: The curated datasets, model scores and antibody synonyms are free to download at https://huggingface.co/datasets/InstaDeepAI/cov-unibind. Folded structures are available upon request.

5 November 2025 in bioRxiv
Verified
Preprint
Open access
ABAryan Bhasin
Aryan Bhasin
ADAntoine Delaunay
Antoine Delaunay
FSFrancesco Saccon
Francesco Saccon
YFYunguan Fu
Yunguan Fu
Protein folding models have revolutionized structure prediction but struggle to capture conformational flexibility. Recent studies perturb inputs or parameters to sample alternative conformations, while diffusion-based approaches generate conformational ensembles directly. Although the former have been benchmarked to some extent, the latter have yet to be evaluated, and sub-domain dynamics validation remains limited. Here, we present a systematic benchmark of nine methods across 20 monomeric proteins with active and inactive states. We extend the pairwise aligned error metric to ensembles and reveal that protein identity exerts a non-negligible influence on model performance. Focusing on Adenylate Kinase, a well-studied enzyme with extensive molecular dynamics (MD) data, we find that Chai-1 performs the best in recovering known conformations, identifying mobile regions, and capturing transition trajectories. These results highlight the potential of generative models as efficient alternatives to MD for exploring protein conformational dynamics and provide a rigorous benchmark for dynamic structure prediction.

11 September 2026 in bioRxiv
Verified
Preprint
Open access
PDF file
QTQuentin Trolliet
Quentin Trolliet
AAAlex Abrudan
Alex Abrudan
ABAryan Bhasin
Aryan Bhasin
YFYunguan Fu
Yunguan Fu
JEJoao Paulo Euko
Joao Paulo Euko
FSFrancesco Saccon
Francesco Saccon
Peptides occupy a valuable niche between small molecules and biologics, but the clinical translation of de novo peptide designs requires rigorous scoring to simultaneously optimise target binding affinity alongside multiple developability traits, including stability, membrane permeability, aggregation propensity, and non-fouling behaviour. Here, we evaluate two distinct approaches for scoring these candidates: classical chemical descriptors and modern deep learning representations derived from protein language and folding models. Assembling nine public datasets spanning five developability traits and four binding-affinity endpoints, we find sequence-derived chemical descriptors alone contain sufficient information to predict developability task labels effectively. Given their drastically lower computational cost and higher interpretability, classical machine learning models trained on these simple descriptors frequently match or approach the performance of complex deep learning architectures, emerging as a highly efficient and interpretable alternative for high-throughput scoring. Finally, for scoring binding affinity, we demonstrate that Boltz-2 pair representations capture the most information among the tested representations; however, the model's predictive power is confounded by a significant bias from the molecular weight of the peptides. Together, these results establish a comprehensive assessment of state-of-the-art methods for predicting both peptide developability and binding affinity, highlighting the enduring value of interpretable chemical descriptors alongside deep learning in the scoring and selection of de novo peptide designs.