Vijil Chenthamarakshan
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Vijil Chenthamarakshan has a total of 67 co-authors. Here are the top co-authors by number of shared publications:
- Payel Das
IBM Research Yorktown Heights NY United States
9 shared publications - Charlotte Mary Deane
Oxford Protein Informatics Group, Department of Statistics, University of Oxford, Oxford OX1 3LB, UK
4 shared publications Inkit Padhi
IBM Research - Yorktown Heights, New York, USA
3 shared publications- Lucy Vost
Department of Statistics, University of Oxford, Oxford, UK
3 shared publications - Youssef Mroueh
IBM Research, Yorktown Heights, NY 10598, USA
3 shared publications - Anna Carbery
University of Oxford, Diamond Light Source
2 shared publications
Latest Publications
- 12 May 2026
- Published by American Association for the Advancement of Science (AAAS) in Health Data Science
- Vol. 6, 0454
BioMamba: Domain-Adaptive Biomedical Language Models
Background: Biomedical language models should improve performance on biomedical text while retaining general-language-modeling fluency. For Mamba-based models, this trade-off has not been systematically studied across biomedical literature and clinical text. Methods: We developed BioMamba, a family of biomedical Mamba2 models at 5 scales (130M, 370M, 780M, 1.3B, and 2.7B) obtained by continued pretraining of released public Mamba2 checkpoints on a balanced 80%/10%/10% mixture of PubMed abstracts, the Colossal Clean Crawled Corpus (C4), and Wikipedia. The architecture is unchanged from Mamba2; the contribution is the adaptation recipe and the accompanying open-weight checkpoints. We evaluated internal language modeling on fixed held-out sets, out-of-domain multiple-choice benchmarks, and 3 downstream tasks across multiple model scales: clinical note completion and discharge summary generation on MIMIC-IV-Note, and biomedical yes/no question answering on Biomedical Semantic Indexing and Question Answering (BioASQ) and PubMed Question Answering (PubMedQA). Results: Across 5 scales, BioMamba consistently lowered PubMed perplexity, improved Wikipedia-style held-out perplexity by 1.46 to 4.72 PPL, and left C4 perplexity essentially unchanged ( ∣ Δ P P L ∣ ≤ 0 .17 ). On 6 out-of-domain multiple-choice benchmarks, BioMamba stayed within ±3 pp of Mamba2 with no systematic regression. After supervised fine-tuning, BioMamba+SFT matched or exceeded Mamba2+SFT on MIMIC-IV note completion and discharge summary generation at every evaluated scale (paired-bootstrap P Δ > 0 ≥ 0 .997 at 2.7B) and improved PubMedQA at every scale. The strongest model (BioMamba-2.7B) reached a PubMed perplexity of 5.28 and accuracies of 90.24% and 73.00% on BioASQ and PubMedQA, respectively. Conclusions: A balanced domain-adaptive continued pretraining recipe strengthens Mamba2 language models on biomedical literature and clinical text while preserving general-language-modeling fluency.
- 12 May 2026
- Published by American Association for the Advancement of Science (AAAS) in Health Data Science
- Vol. 6, 0454
BioMamba: Domain-Adaptive Biomedical Language Models
Background: Biomedical language models should improve performance on biomedical text while retaining general-language-modeling fluency. For Mamba-based models, this trade-off has not been systematically studied across biomedical literature and clinical text. Methods: We developed BioMamba, a family of biomedical Mamba2 models at 5 scales (130M, 370M, 780M, 1.3B, and 2.7B) obtained by continued pretraining of released public Mamba2 checkpoints on a balanced 80%/10%/10% mixture of PubMed abstracts, the Colossal Clean Crawled Corpus (C4), and Wikipedia. The architecture is unchanged from Mamba2; the contribution is the adaptation recipe and the accompanying open-weight checkpoints. We evaluated internal language modeling on fixed held-out sets, out-of-domain multiple-choice benchmarks, and 3 downstream tasks across multiple model scales: clinical note completion and discharge summary generation on MIMIC-IV-Note, and biomedical yes/no question answering on Biomedical Semantic Indexing and Question Answering (BioASQ) and PubMed Question Answering (PubMedQA). Results: Across 5 scales, BioMamba consistently lowered PubMed perplexity, improved Wikipedia-style held-out perplexity by 1.46 to 4.72 PPL, and left C4 perplexity essentially unchanged ( ∣ Δ P P L ∣ ≤ 0 .17 ). On 6 out-of-domain multiple-choice benchmarks, BioMamba stayed within ±3 pp of Mamba2 with no systematic regression. After supervised fine-tuning, BioMamba+SFT matched or exceeded Mamba2+SFT on MIMIC-IV note completion and discharge summary generation at every evaluated scale (paired-bootstrap P Δ > 0 ≥ 0 .997 at 2.7B) and improved PubMedQA at every scale. The strongest model (BioMamba-2.7B) reached a PubMed perplexity of 5.28 and accuracies of 90.24% and 73.00% on BioASQ and PubMedQA, respectively. Conclusions: A balanced domain-adaptive continued pretraining recipe strengthens Mamba2 language models on biomedical literature and clinical text while preserving general-language-modeling fluency.
- Open Access
- 18 August 2025
- Published by Royal Society of Chemistry (RSC) in Digital Discovery
- Vol. 4 (10) , 2684-2696
GP-MoLFormer: a foundation model for molecular generation
GP-MoLFormer, a GPT-style chemical foundation model trained on 0.65–1.1b SMILES, reveals the impact of training and inference scaling on generation quality. With pair-tuning, a novel fine-tuning method, it enables efficient molecular optimization.
- 18 August 2025
- Published by Royal Society of Chemistry (RSC) in Digital Discovery
- Vol. 4 (10) , 2684-2696
Open AccessGP-MoLFormer: a foundation model for molecular generation
GP-MoLFormer, a GPT-style chemical foundation model trained on 0.65–1.1b SMILES, reveals the impact of training and inference scaling on generation quality. With pair-tuning, a novel fine-tuning method, it enables efficient molecular optimization.
- Open Access
- 24 March 2025
- Published by Royal Society of Chemistry (RSC) in Digital Discovery
- Vol. 4 (4) , 1092-1099
Improving structural plausibility in diffusion-based 3D molecule generation via property-conditioned training with distorted molecules
Conditional diffusion models improve molecular generation by learning from both high- and low-quality structures, using quality-labeled distorted molecules to enhance plausibility and enable selective sampling.
- 24 March 2025
- Published by Royal Society of Chemistry (RSC) in Digital Discovery
- Vol. 4 (4) , 1092-1099
Open AccessImproving structural plausibility in diffusion-based 3D molecule generation via property-conditioned training with distorted molecules
Conditional diffusion models improve molecular generation by learning from both high- and low-quality structures, using quality-labeled distorted molecules to enhance plausibility and enable selective sampling.
- Open Access
Improving Structural Plausibility in 3D Molecule Generation via Property-Conditioned Training with Distorted Molecules
Traditional drug design methods are costly and time-consuming due to their reliance on trial-and-error processes. As a result, computational methods, including diffusion models, designed for molecule generation tasks have gained significant traction. Despite their potential, they have faced criticism for producing physically implausible outputs. We alleviate this problem by conditionally training a diffusion model capable of generating molecules of varying and controllable levels of structural plausibility. This is achieved by adding distorted molecules to training datasets, and then annotating each molecule with a label representing the extent of its distortion, and hence its quality. By training the model to distinguish between favourable and unfavourable molecular conformations alongside the standard molecule generation training process, we can selectively sample molecules from the high-quality region of learned space, resulting in improvements in the validity of generated molecules. In addition to the standard two datasets used by molecule generation methods (QM9 and GEOM), we also test our method on a druglike dataset derived from ZINC. We use our conditional method with EDM, the first E(3) equivariant diffusion model for molecule generation, as well as two further models—a more recent diffusion model and a flow matching model—which were built off EDM. We demonstrate improvements in validity as assessed by RD-Kit parsability and the PoseBusters test suite; more broadly, though, our findings highlight the effectiveness of conditioning methods on low-quality data to improve the sampling of high-quality data.
Open AccessImproving Structural Plausibility in 3D Molecule Generation via Property-Conditioned Training with Distorted Molecules
Traditional drug design methods are costly and time-consuming due to their reliance on trial-and-error processes. As a result, computational methods, including diffusion models, designed for molecule generation tasks have gained significant traction. Despite their potential, they have faced criticism for producing physically implausible outputs. We alleviate this problem by conditionally training a diffusion model capable of generating molecules of varying and controllable levels of structural plausibility. This is achieved by adding distorted molecules to training datasets, and then annotating each molecule with a label representing the extent of its distortion, and hence its quality. By training the model to distinguish between favourable and unfavourable molecular conformations alongside the standard molecule generation training process, we can selectively sample molecules from the high-quality region of learned space, resulting in improvements in the validity of generated molecules. In addition to the standard two datasets used by molecule generation methods (QM9 and GEOM), we also test our method on a druglike dataset derived from ZINC. We use our conditional method with EDM, the first E(3) equivariant diffusion model for molecule generation, as well as two further models—a more recent diffusion model and a flow matching model—which were built off EDM. We demonstrate improvements in validity as assessed by RD-Kit parsability and the PoseBusters test suite; more broadly, though, our findings highlight the effectiveness of conditioning methods on low-quality data to improve the sampling of high-quality data.
- Top Cited PapersOpen Access
- 23 June 2023
- Published by Springer Nature in Communications Chemistry
- Vol. 6 (1) , 1-12
An end-to-end deep learning framework for translating mass spectra to de-novo molecules
Elucidating the structure of a chemical compound is a fundamental task in chemistry with applications in multiple domains including drug discovery, precision medicine, and biomarker discovery. The common practice for elucidating the structure of a compound is to obtain a mass spectrum and subsequently retrieve its structure from spectral databases. However, these methods fail for novel molecules that are not present in the reference database. We propose Spec2Mol, a deep learning architecture for molecular structure recommendation given mass spectra alone. Spec2Mol is inspired by the Speech2Text deep learning architectures for translating audio signals into text. Our approach is based on an encoder-decoder architecture. The encoder learns the spectra embeddings, while the decoder, pre-trained on a massive dataset of chemical structures for translating between different molecular representations, reconstructs SMILES sequences of the recommended chemical structures. We have evaluated Spec2Mol by assessing the molecular similarity between the recommended structures and the original structure. Our analysis showed that Spec2Mol is able to identify the presence of key molecular substructures from its mass spectrum, and shows on par performance, when compared to existing fragmentation tree methods particularly when test structure information is not available during training or present in the reference database.
- 23 June 2023
- Published by Springer Nature in Communications Chemistry
- Vol. 6 (1) , 1-12
Top Cited PapersOpen AccessAn end-to-end deep learning framework for translating mass spectra to de-novo molecules
Elucidating the structure of a chemical compound is a fundamental task in chemistry with applications in multiple domains including drug discovery, precision medicine, and biomarker discovery. The common practice for elucidating the structure of a compound is to obtain a mass spectrum and subsequently retrieve its structure from spectral databases. However, these methods fail for novel molecules that are not present in the reference database. We propose Spec2Mol, a deep learning architecture for molecular structure recommendation given mass spectra alone. Spec2Mol is inspired by the Speech2Text deep learning architectures for translating audio signals into text. Our approach is based on an encoder-decoder architecture. The encoder learns the spectra embeddings, while the decoder, pre-trained on a massive dataset of chemical structures for translating between different molecular representations, reconstructs SMILES sequences of the recommended chemical structures. We have evaluated Spec2Mol by assessing the molecular similarity between the recommended structures and the original structure. Our analysis showed that Spec2Mol is able to identify the presence of key molecular substructures from its mass spectrum, and shows on par performance, when compared to existing fragmentation tree methods particularly when test structure information is not available during training or present in the reference database.
- Open Access
- 23 June 2023
- Published by American Association for the Advancement of Science (AAAS) in Science Advances
- Vol. 9 (25) , eadg7865
Accelerating drug target inhibitor discovery with a deep generative foundation model
Inhibitor discovery for emerging drug-target proteins is challenging, especially when target structure or active molecules are unknown. Here, we experimentally validate the broad utility of a deep generative framework trained at-scale on protein sequences, small molecules, and their mutual interactions—unbiased toward any specific target. We performed a protein sequence-conditioned sampling on the generative foundation model to design small-molecule inhibitors for two dissimilar targets: the spike protein receptor-binding domain (RBD) and the main protease from SARS-CoV-2. Despite using only the target sequence information during the model inference, micromolar-level inhibition was observed in vitro for two candidates out of four synthesized for each target. The most potent spike RBD inhibitor exhibited activity against several variants in live virus neutralization assays. These results establish that a single, broadly deployable generative foundation model for accelerated inhibitor discovery is effective and efficient, even in the absence of target structure or binder information.
- 23 June 2023
- Published by American Association for the Advancement of Science (AAAS) in Science Advances
- Vol. 9 (25) , eadg7865
Open AccessAccelerating drug target inhibitor discovery with a deep generative foundation model
Inhibitor discovery for emerging drug-target proteins is challenging, especially when target structure or active molecules are unknown. Here, we experimentally validate the broad utility of a deep generative framework trained at-scale on protein sequences, small molecules, and their mutual interactions—unbiased toward any specific target. We performed a protein sequence-conditioned sampling on the generative foundation model to design small-molecule inhibitors for two dissimilar targets: the spike protein receptor-binding domain (RBD) and the main protease from SARS-CoV-2. Despite using only the target sequence information during the model inference, micromolar-level inhibition was observed in vitro for two candidates out of four synthesized for each target. The most potent spike RBD inhibitor exhibited activity against several variants in live virus neutralization assays. These results establish that a single, broadly deployable generative foundation model for accelerated inhibitor discovery is effective and efficient, even in the absence of target structure or binder information.
- Open Access
- 11 May 2023
- Published by American Chemical Society (ACS) in Journal of Chemical Information and Modeling
- Vol. 63 (10) , 2960-2974
A Small Step Toward Generalizability: Training a Machine Learning Scoring Function for Structure-Based Virtual Screening
Over the past few years, many machine learning-based scoring functions for predicting the binding of small molecules to proteins have been developed. Their objective is to approximate the distribution which takes two molecules as input and outputs the energy of their interaction. Only a scoring function that accounts for the interatomic interactions involved in binding can accurately predict binding affinity on unseen molecules. However, many scoring functions make predictions based on data set biases rather than an understanding of the physics of binding. These scoring functions perform well when tested on similar targets to those in the training set but fail to generalize to dissimilar targets. To test what a machine learning-based scoring function has learned, input attribution, a technique for learning which features are important to a model when making a prediction on a particular data point, can be applied. If a model successfully learns something beyond data set biases, attribution should give insight into the important binding interactions that are taking place. We built a machine learning-based scoring function that aimed to avoid the influence of bias via thorough train and test data set filtering and show that it achieves comparable performance on the Comparative Assessment of Scoring Functions, 2016 (CASF-2016) benchmark to other leading methods. We then use the CASF-2016 test set to perform attribution and find that the bonds identified as important by PointVS, unlike those extracted from other scoring functions, have a high correlation with those found by a distance-based interaction profiler. We then show that attribution can be used to extract important binding pharmacophores from a given protein target when supplied with a number of bound structures. We use this information to perform fragment elaboration and see improvements in docking scores compared to using structural information from a traditional, data-based approach. This not only provides definitive proof that the scoring function has learned to identify some important binding interactions but also constitutes the first deep learning-based method for extracting structural information from a target for molecule design.
- 11 May 2023
- Published by American Chemical Society (ACS) in Journal of Chemical Information and Modeling
- Vol. 63 (10) , 2960-2974
Open AccessA Small Step Toward Generalizability: Training a Machine Learning Scoring Function for Structure-Based Virtual Screening
Over the past few years, many machine learning-based scoring functions for predicting the binding of small molecules to proteins have been developed. Their objective is to approximate the distribution which takes two molecules as input and outputs the energy of their interaction. Only a scoring function that accounts for the interatomic interactions involved in binding can accurately predict binding affinity on unseen molecules. However, many scoring functions make predictions based on data set biases rather than an understanding of the physics of binding. These scoring functions perform well when tested on similar targets to those in the training set but fail to generalize to dissimilar targets. To test what a machine learning-based scoring function has learned, input attribution, a technique for learning which features are important to a model when making a prediction on a particular data point, can be applied. If a model successfully learns something beyond data set biases, attribution should give insight into the important binding interactions that are taking place. We built a machine learning-based scoring function that aimed to avoid the influence of bias via thorough train and test data set filtering and show that it achieves comparable performance on the Comparative Assessment of Scoring Functions, 2016 (CASF-2016) benchmark to other leading methods. We then use the CASF-2016 test set to perform attribution and find that the bonds identified as important by PointVS, unlike those extracted from other scoring functions, have a high correlation with those found by a distance-based interaction profiler. We then show that attribution can be used to extract important binding pharmacophores from a given protein target when supplied with a number of bound structures. We use this information to perform fragment elaboration and see improvements in docking scores compared to using structural information from a traditional, data-based approach. This not only provides definitive proof that the scoring function has learned to identify some important binding interactions but also constitutes the first deep learning-based method for extracting structural information from a target for molecule design.
- Open Access
- 1 May 2023
- Published by Springer Nature in npj Computational Materials
- Vol. 9 (1) , 1-6
Accelerating material design with the generative toolkit for scientific discovery
With the growing availability of data within various scientific domains, generative models hold enormous potential to accelerate scientific discovery. They harness powerful representations learned from datasets to speed up the formulation of novel hypotheses with the potential to impact material discovery broadly. We present the Generative Toolkit for Scientific Discovery (GT4SD). This extensible open-source library enables scientists, developers, and researchers to train and use state-of-the-art generative models to accelerate scientific discovery focused on organic material design.
- 1 May 2023
- Published by Springer Nature in npj Computational Materials
- Vol. 9 (1) , 1-6
Open AccessAccelerating material design with the generative toolkit for scientific discovery
With the growing availability of data within various scientific domains, generative models hold enormous potential to accelerate scientific discovery. They harness powerful representations learned from datasets to speed up the formulation of novel hypotheses with the potential to impact material discovery broadly. We present the Generative Toolkit for Scientific Discovery (GT4SD). This extensible open-source library enables scientists, developers, and researchers to train and use state-of-the-art generative models to accelerate scientific discovery focused on organic material design.
- Top Cited Papers
- 21 December 2022
- Published by Springer Nature in Nature Machine Intelligence
- Vol. 4 (12) , 1256-1264
Large-scale chemical language representations capture molecular structure and properties
- 21 December 2022
- Published by Springer Nature in Nature Machine Intelligence
- Vol. 4 (12) , 1256-1264
Top Cited PapersLarge-scale chemical language representations capture molecular structure and properties
- Open Access
A Step Towards Generalisability: Training a Machine Learning Scoring Function for Structure-Based Virtual Screening
Over the last few years, many machine learning-based scoring functions for predicting the binding of small molecules to proteins have been developed. Their objective is to approximate the distribution which takes two molecules as input and outputs the energy of their interaction. Only a scoring function that accounts for the interatomic interactions involved in binding can accurately predict binding affinity on unseen molecules. However, many scoring functions make predictions based on dataset biases rather than an understanding of the physics of binding. These scoring functions perform well when tested on similar targets to those in the training set, but fail to generalise to dissimilar targets. To test what a machine learning-based scoring function has learnt, input attribution—a technique for learning which features are important to a model when making a prediction on a particular data point—can be applied. If a model successfully learns something beyond dataset biases, attribution should give insight into the important binding interactions that are taking place. We built a machine learning-based scoring function that aimed to avoid the influence of bias via thorough train and test dataset filtering, and show that it achieves comparable performance on the CASF-2016 benchmark to other leading methods. We then use the CASF-2016 test set to perform attribution, and find that the bonds identified as important by PointVS, unlike those extracted from other scoring functions, have a high correlation with those found by a distance-based interaction profiler. We then show that attribution can be used to extract important binding pharmacophores from a given protein target when supplied with a number of bound structures. We use this information to perform fragment elaboration, and see improvements in docking scores compared to using structural information from a traditional, data-based approach. This not only provides definitive proof that the scoring function has learnt to identify some important binding interactions, but also constitutes the first deep learning-based method for extracting structural information from a target for molecule design.
Open AccessA Step Towards Generalisability: Training a Machine Learning Scoring Function for Structure-Based Virtual Screening
Over the last few years, many machine learning-based scoring functions for predicting the binding of small molecules to proteins have been developed. Their objective is to approximate the distribution which takes two molecules as input and outputs the energy of their interaction. Only a scoring function that accounts for the interatomic interactions involved in binding can accurately predict binding affinity on unseen molecules. However, many scoring functions make predictions based on dataset biases rather than an understanding of the physics of binding. These scoring functions perform well when tested on similar targets to those in the training set, but fail to generalise to dissimilar targets. To test what a machine learning-based scoring function has learnt, input attribution—a technique for learning which features are important to a model when making a prediction on a particular data point—can be applied. If a model successfully learns something beyond dataset biases, attribution should give insight into the important binding interactions that are taking place. We built a machine learning-based scoring function that aimed to avoid the influence of bias via thorough train and test dataset filtering, and show that it achieves comparable performance on the CASF-2016 benchmark to other leading methods. We then use the CASF-2016 test set to perform attribution, and find that the bonds identified as important by PointVS, unlike those extracted from other scoring functions, have a high correlation with those found by a distance-based interaction profiler. We then show that attribution can be used to extract important binding pharmacophores from a given protein target when supplied with a number of bound structures. We use this information to perform fragment elaboration, and see improvements in docking scores compared to using structural information from a traditional, data-based approach. This not only provides definitive proof that the scoring function has learnt to identify some important binding interactions, but also constitutes the first deep learning-based method for extracting structural information from a target for molecule design.