Sanjana Tule
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Sanjana Tule has a total of 23 co-authors. Here are the top co-authors by number of shared publications:
- Mikael Bodén
The University of Queensland
8 shared publications - Gabriel Foley
School of Chemistry and Molecular Biosciences, The University of Queensland
4 shared publications - Chongting Zhao
School of Chemistry and Molecular Biosciences, The University of Queensland , Brisbane, QLD 4072, Australia
2 shared publications - Michael Forbes
School of Mathematics and Physics, The University of Queensland, St. Lucia, Queensland 4072, Australia
2 shared publications - Apoorva Prabhu
School of Chemistry and Molecular Biosciences, Australian Centre for Ecogenomics, The University of Queensland, Brisbane, Queensland, Australia
1 shared publication Ariane Mora
1 shared publication
Latest Publications
- Open Access
- 16 July 2026
- Published by Elsevier in SSRN Electronic Journal
Generating Hybrid Proteins with the MSA-Transformer
Protein superfamilies encompass substantial sequence and functional diversity, with desirable properties distributed across different family members. Efforts to marry these properties by recombination of homologous members tend to disrupt their integrity pointing to a challenging protein design search problem. We introduce a framework to use the MSA-Transformer in a stochastic iterative procedure to generate hybrid proteins between homologous proteins. We demonstrate that when applied across multiple sequence alignments (MSAs) of diverse protein families it can produce sequences that occupy biologically meaningful regions of sequence space; according to in silico metrics, sequences achieve higher consistency and plausibility scores than naive baseline searches. In the B1/B2 metallo-beta-lactamase family, generated hybrids largely retain the conserved core fold while recombining structural and active-site motifs from both subclasses, demonstrating the model’s capacity to preserve catalytic features while exploring novel structural permutations. These results suggest that protein language models trained on multiple sequence alignments can facilitate protein engineering by guiding residue-level recombination within homologous families.
- 16 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessGenerating Hybrid Proteins with the MSA-Transformer
Protein superfamilies encompass substantial sequence and functional diversity, with desirable properties distributed across different family members. Efforts to marry these properties by recombination of homologous members tend to disrupt their integrity pointing to a challenging protein design search problem. We introduce a framework to use the MSA-Transformer in a stochastic iterative procedure to generate hybrid proteins between homologous proteins. We demonstrate that when applied across multiple sequence alignments (MSAs) of diverse protein families it can produce sequences that occupy biologically meaningful regions of sequence space; according to in silico metrics, sequences achieve higher consistency and plausibility scores than naive baseline searches. In the B1/B2 metallo-beta-lactamase family, generated hybrids largely retain the conserved core fold while recombining structural and active-site motifs from both subclasses, demonstrating the model’s capacity to preserve catalytic features while exploring novel structural permutations. These results suggest that protein language models trained on multiple sequence alignments can facilitate protein engineering by guiding residue-level recombination within homologous families.
- Open Access
Generating Hybrid Proteins with the MSA-Transformer
Protein superfamilies display extensive sequence and functional divergence, providing a rich landscape for engineering functionally enhanced variants. We present a stochastic, iterative framework that leverages the MSA-Transformer to generate intermediate sequences between a homologous pair of user-specified “source” and “target” proteins in sequence space. Sequence sites to “mask” is either selected by embedding-based dissimilarity or by row-attention information, while beam search concurrently explores multiple mutational pathways. Pre-trained sparse autoencoders combined with sequence and structural analyses are used to trace the inheritance and exchange of features across the “mutational pathway”, revealing “hybrid” sequences that integrate properties of both source and target proteins. Applied across diverse protein families, the framework produces sequences that occupy biologically meaningful regions of sequence space and achieve higher consistency and plausibility scores than random baselines according to in silico metrics. In the B1/B2 metallo- β -lactamase family, hybrids largely retain their core fold recombining structural and active-site motifs from both subclasses, demonstrating the model’s capacity to preserve catalytic features while exploring novel structural permutations. Code availability: The implementation is available via GitHub at https://github.com/santule/protmixy .
Open AccessGenerating Hybrid Proteins with the MSA-Transformer
Protein superfamilies display extensive sequence and functional divergence, providing a rich landscape for engineering functionally enhanced variants. We present a stochastic, iterative framework that leverages the MSA-Transformer to generate intermediate sequences between a homologous pair of user-specified “source” and “target” proteins in sequence space. Sequence sites to “mask” is either selected by embedding-based dissimilarity or by row-attention information, while beam search concurrently explores multiple mutational pathways. Pre-trained sparse autoencoders combined with sequence and structural analyses are used to trace the inheritance and exchange of features across the “mutational pathway”, revealing “hybrid” sequences that integrate properties of both source and target proteins. Applied across diverse protein families, the framework produces sequences that occupy biologically meaningful regions of sequence space and achieve higher consistency and plausibility scores than random baselines according to in silico metrics. In the B1/B2 metallo- β -lactamase family, hybrids largely retain their core fold recombining structural and active-site motifs from both subclasses, demonstrating the model’s capacity to preserve catalytic features while exploring novel structural permutations. Code availability: The implementation is available via GitHub at https://github.com/santule/protmixy .
- Open Access
Comprehensive molecular impact mapping of common and rare variants at GWAS loci
Deep learning sequence to function models can predict the molecular effects of genetic variants, but their predictions are limited to the cell types and assays they are trained on. Here we describe DNACipher, a deep learning model that predicts the effects of genetic variants across diverse biological contexts-including those not directly measured. DNACipher takes 196 kb of genome sequences as input and imputes variant effects across 38,582 cell type-assay combinations. DNACipher generates predictions for >7 times as many contexts as Enformer, which allows for better detection of variant effects at expression quantitative trait loci (eQTLs). We also introduce DNACipher Deep Variant Impact Mapping (DVIM), a method to identify variants with molecular effects at genome-wide association study (GWAS) loci. Application of DVIM to type 1 diabetes (T1D) reduced the mean fine-mapping credible set size from 24 to 1.4 variants per signal. DVIM variants had significantly higher fine-mapping posterior probabilities, and their predicted effects were supported by single-nucleus ATAC-seq and luciferase assays. DVIM also detected 6547 rare variants with molecular effects at 96% of GWAS T1D loci, and these were enriched for associations with immune traits. In summary, DNACipher DVIM prioritises common and rare variants at GWAS loci by predicting molecular effects across a broad range of contexts.
Open AccessComprehensive molecular impact mapping of common and rare variants at GWAS loci
Deep learning sequence to function models can predict the molecular effects of genetic variants, but their predictions are limited to the cell types and assays they are trained on. Here we describe DNACipher, a deep learning model that predicts the effects of genetic variants across diverse biological contexts-including those not directly measured. DNACipher takes 196 kb of genome sequences as input and imputes variant effects across 38,582 cell type-assay combinations. DNACipher generates predictions for >7 times as many contexts as Enformer, which allows for better detection of variant effects at expression quantitative trait loci (eQTLs). We also introduce DNACipher Deep Variant Impact Mapping (DVIM), a method to identify variants with molecular effects at genome-wide association study (GWAS) loci. Application of DVIM to type 1 diabetes (T1D) reduced the mean fine-mapping credible set size from 24 to 1.4 variants per signal. DVIM variants had significantly higher fine-mapping posterior probabilities, and their predicted effects were supported by single-nucleus ATAC-seq and luciferase assays. DVIM also detected 6547 rare variants with molecular effects at 96% of GWAS T1D loci, and these were enriched for associations with immune traits. In summary, DNACipher DVIM prioritises common and rare variants at GWAS loci by predicting molecular effects across a broad range of contexts.
- Open Access
- 22 November 2024
- Published by Oxford University Press (OUP) in Briefings in Bioinformatics
- Vol. 26 (1)
Do protein language models learn phylogeny?
Deep machine learning demonstrates a capacity to uncover evolutionary relationships directly from protein sequences, in effect internalising notions inherent to classical phylogenetic tree inference. We connect these two paradigms by assessing the capacity of protein-based language models (pLMs) to discern phylogenetic relationships without being explicitly trained to do so. We evaluate ESM2, ProtTrans, and MSA-Transformer relative to classical phylogenetic methods, while also considering sequence insertions and deletions (indels) across 114 Pfam datasets. The largest ESM2 model tends to outperform other pLMs (including the multimodal ESM3) by recovering phylogenetic relationships among homologous protein sequences in both low- and high-gap settings. pLMs agree with conventional phylogenetic methods in general, but more so for protein families with fewer implied indels, highlighting indels as a key factor differentiating classical phylogenetics from pLMs. We find that pLMs preferentially capture broader as opposed to finer evolutionary relationships within a specific protein family, where ESM2 has a sweet spot for highly divergent sequences, at remote distance. Less than 10% of neurons are sufficient to broadly recapitulate classical phylogenetic distances; when used in isolation, the difference between the paradigms is further diminished. We show these neurons are polysemantic, shared among different homologous families but never fully overlapping. We highlight the potential of ESM2 as a complementary tool for phylogenetic analysis, especially when extending to remote homologs that are difficult to align and imply complex histories of insertions and deletions. Implementations of analyses are available at https://github.com/santule/pLMEvo.
- 22 November 2024
- Published by Oxford University Press (OUP) in Briefings in Bioinformatics
- Vol. 26 (1)
Open AccessDo protein language models learn phylogeny?
Deep machine learning demonstrates a capacity to uncover evolutionary relationships directly from protein sequences, in effect internalising notions inherent to classical phylogenetic tree inference. We connect these two paradigms by assessing the capacity of protein-based language models (pLMs) to discern phylogenetic relationships without being explicitly trained to do so. We evaluate ESM2, ProtTrans, and MSA-Transformer relative to classical phylogenetic methods, while also considering sequence insertions and deletions (indels) across 114 Pfam datasets. The largest ESM2 model tends to outperform other pLMs (including the multimodal ESM3) by recovering phylogenetic relationships among homologous protein sequences in both low- and high-gap settings. pLMs agree with conventional phylogenetic methods in general, but more so for protein families with fewer implied indels, highlighting indels as a key factor differentiating classical phylogenetics from pLMs. We find that pLMs preferentially capture broader as opposed to finer evolutionary relationships within a specific protein family, where ESM2 has a sweet spot for highly divergent sequences, at remote distance. Less than 10% of neurons are sufficient to broadly recapitulate classical phylogenetic distances; when used in isolation, the difference between the paradigms is further diminished. We show these neurons are polysemantic, shared among different homologous families but never fully overlapping. We highlight the potential of ESM2 as a complementary tool for phylogenetic analysis, especially when extending to remote homologs that are difficult to align and imply complex histories of insertions and deletions. Implementations of analyses are available at https://github.com/santule/pLMEvo.
- Open Access
Do Protein Language Models Learn Phylogeny?
Deep machine learning demonstrates a capacity to uncover evolutionary relationships directly from protein sequences, in effect internalising notions inherent to classical phylogenetic tree inference. We connect these two paradigms by assessing the capacity of protein-based language models (pLMs) to discern phylogenetic relationships without being explicitly trained to do so. We evaluate ESM2, ProtTrans and MSA-Transformer relative to classical phylogenetic methods, while also considering sequence insertions and deletions (indels) across 114 Pfam datasets. The largest ESM2 model tends to outperform other pLMs (including the multimodal ESM3) by recovering phylogenetic relationships among homologous protein sequences in both low- and high-gap settings. pLMs agree with conventional phylogenetic methods in general, but more so for protein families with fewer implied indels, highlighting indels as a key factor differentiating classical phylogenetics from pLMs. We find that pLMs preferentially capture broader as opposed to finer evolutionary relationships within a specific protein family, where ESM2 has a sweet spot for highly divergent sequences, at remote distance. Less than 10% of neurons are sufficient to broadly recapitulate classical phylogenetic distances; when used in isolation the difference between the paradigms is further diminished. We show these neurons are polysemantic, shared among different homologous families but never fully overlapping. We highlight the potential of ESM2 as a complementary tool for phylogenetic analysis, especially when extending to remote homologs that are difficult to align and imply complex histories of insertions and deletions.
Open AccessDo Protein Language Models Learn Phylogeny?
Deep machine learning demonstrates a capacity to uncover evolutionary relationships directly from protein sequences, in effect internalising notions inherent to classical phylogenetic tree inference. We connect these two paradigms by assessing the capacity of protein-based language models (pLMs) to discern phylogenetic relationships without being explicitly trained to do so. We evaluate ESM2, ProtTrans and MSA-Transformer relative to classical phylogenetic methods, while also considering sequence insertions and deletions (indels) across 114 Pfam datasets. The largest ESM2 model tends to outperform other pLMs (including the multimodal ESM3) by recovering phylogenetic relationships among homologous protein sequences in both low- and high-gap settings. pLMs agree with conventional phylogenetic methods in general, but more so for protein families with fewer implied indels, highlighting indels as a key factor differentiating classical phylogenetics from pLMs. We find that pLMs preferentially capture broader as opposed to finer evolutionary relationships within a specific protein family, where ESM2 has a sweet spot for highly divergent sequences, at remote distance. Less than 10% of neurons are sufficient to broadly recapitulate classical phylogenetic distances; when used in isolation the difference between the paradigms is further diminished. We show these neurons are polysemantic, shared among different homologous families but never fully overlapping. We highlight the potential of ESM2 as a complementary tool for phylogenetic analysis, especially when extending to remote homologs that are difficult to align and imply complex histories of insertions and deletions.
- Open Access
- 28 June 2024
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 40 (Supplement) , i277-i286
Optimal phylogenetic reconstruction of insertion and deletion events
Motivation: Insertions and deletions (indels) influence the genetic code in fundamentally distinct ways from substitutions, significantly impacting gene product structure and function. Despite their influence, the evolutionary history of indels is often neglected in phylogenetic tree inference and ancestral sequence reconstruction, hindering efforts to comprehend biological diversity determinants and engineer variants for medical and industrial applications. Results: We frame determining the optimal history of indel events as a single Mixed-Integer Programming (MIP) problem, across all branch points in a phylogenetic tree adhering to topological constraints, and all sites implied by a given set of aligned, extant sequences. By disentangling the impact on ancestral sequences at each branch point, this approach identifies the minimal indel events that jointly explain the diversity in sequences mapped to the tips of that tree. MIP can recover alternate optimal indel histories, if available. We evaluated MIP for indel inference on a dataset comprising 15 real phylogenetic trees associated with protein families ranging from 165 to 2000 extant sequences, and on 60 synthetic trees at comparable scales of data and reflecting realistic rates of mutation. Across relevant metrics, MIP outperformed alternative parsimony-based approaches and reported the fewest indel events, on par or below their occurrence in synthetic datasets. MIP offers a rational justification for indel patterns in extant sequences; importantly, it uniquely identifies global optima on complex protein data sets without making unrealistic assumptions of independence or evolutionary underpinnings, promising a deeper understanding of molecular evolution and aiding novel protein design. Availability and implementation: The implementation is available via GitHub at https://github.com/santule/indelmip.
- 28 June 2024
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 40 (Supplement) , i277-i286
Open AccessOptimal phylogenetic reconstruction of insertion and deletion events
Motivation: Insertions and deletions (indels) influence the genetic code in fundamentally distinct ways from substitutions, significantly impacting gene product structure and function. Despite their influence, the evolutionary history of indels is often neglected in phylogenetic tree inference and ancestral sequence reconstruction, hindering efforts to comprehend biological diversity determinants and engineer variants for medical and industrial applications. Results: We frame determining the optimal history of indel events as a single Mixed-Integer Programming (MIP) problem, across all branch points in a phylogenetic tree adhering to topological constraints, and all sites implied by a given set of aligned, extant sequences. By disentangling the impact on ancestral sequences at each branch point, this approach identifies the minimal indel events that jointly explain the diversity in sequences mapped to the tips of that tree. MIP can recover alternate optimal indel histories, if available. We evaluated MIP for indel inference on a dataset comprising 15 real phylogenetic trees associated with protein families ranging from 165 to 2000 extant sequences, and on 60 synthetic trees at comparable scales of data and reflecting realistic rates of mutation. Across relevant metrics, MIP outperformed alternative parsimony-based approaches and reported the fewest indel events, on par or below their occurrence in synthetic datasets. MIP offers a rational justification for indel patterns in extant sequences; importantly, it uniquely identifies global optima on complex protein data sets without making unrealistic assumptions of independence or evolutionary underpinnings, promising a deeper understanding of molecular evolution and aiding novel protein design. Availability and implementation: The implementation is available via GitHub at https://github.com/santule/indelmip.
- Open Access
Optimal Phylogenetic Reconstruction of Insertion and Deletion Events
Insertions and deletions (indels) influence the genetic code in fundamentally distinct ways from substitutions, significantly impacting gene product structure and function. Despite their influence, the evolutionary history of indels is often neglected in phylogenetic tree inference and ancestral sequence reconstruction, hindering efforts to comprehend biological diversity determinants and engineer variants for medical and industrial applications. We frame determining the optimal history of indel events as a single Mixed-Integer Programming (MIP) problem, across all nodes in a phylogenetic tree adhering to topological constraints, and all sites implied by a given set of aligned, extant sequences. By disentangling the impact on ancestral sequences at each branch point, this approach identifies the minimal indel events that jointly explain the diversity in sequences mapped to the tips of that tree. MIP can recover alternate optimal indel histories, if available. We evaluated MIP for indel inference on a dataset comprising 15 real phylogenetic trees associated with protein families ranging from 165 to 2000 extant sequences, and on 60 synthetic trees at comparable scales of data and reflecting realistic rates of mutation. Across relevant metrics, MIP outperformed alternative parsimony-based approaches and reported the fewest indel events, on par or below their occurrence in synthetic datasets. MIP offers a rational justification for indel patterns in extant sequences; importantly, it uniquely identifies global optima on complex protein data sets without making unrealistic assumptions of independence or evolutionary underpinnings, promising a deeper understanding of molecular evolution and aiding novel protein design.
Open AccessOptimal Phylogenetic Reconstruction of Insertion and Deletion Events
Insertions and deletions (indels) influence the genetic code in fundamentally distinct ways from substitutions, significantly impacting gene product structure and function. Despite their influence, the evolutionary history of indels is often neglected in phylogenetic tree inference and ancestral sequence reconstruction, hindering efforts to comprehend biological diversity determinants and engineer variants for medical and industrial applications. We frame determining the optimal history of indel events as a single Mixed-Integer Programming (MIP) problem, across all nodes in a phylogenetic tree adhering to topological constraints, and all sites implied by a given set of aligned, extant sequences. By disentangling the impact on ancestral sequences at each branch point, this approach identifies the minimal indel events that jointly explain the diversity in sequences mapped to the tips of that tree. MIP can recover alternate optimal indel histories, if available. We evaluated MIP for indel inference on a dataset comprising 15 real phylogenetic trees associated with protein families ranging from 165 to 2000 extant sequences, and on 60 synthetic trees at comparable scales of data and reflecting realistic rates of mutation. Across relevant metrics, MIP outperformed alternative parsimony-based approaches and reported the fewest indel events, on par or below their occurrence in synthetic datasets. MIP offers a rational justification for indel patterns in extant sequences; importantly, it uniquely identifies global optima on complex protein data sets without making unrealistic assumptions of independence or evolutionary underpinnings, promising a deeper understanding of molecular evolution and aiding novel protein design.
- Open Access
- 8 January 2024
- Published by Oxford University Press (OUP) in ISME Communications
- Vol. 4 (1) , ycae067
Machine learning and metagenomics identifies uncharacterized taxa inferred to drive biogeochemical cycles in a subtropical hypereutrophic estuary
Anthropogenic influences have drastically increased nutrient concentrations in many estuaries globally, and microbial communities have adapted to the resulting hypereutrophic ecosystems. However, our knowledge of the dominant microbial taxa and their potential functions in these ecosystems has remained sparse. Here, we study prokaryotic community dynamics in a temporal–spatial dataset, from a subtropical hypereutrophic estuary. Screening 54 water samples across brackish to marine sites revealed that nutrient concentrations and salinity best explained spatial community variations, whereas temperature and dissolved oxygen likely drive seasonal shifts. By combining short and long read sequencing data, we recovered 2,459 metagenome-assembled genomes, proposed new taxon names for previously uncharacterised lineages, and created an extensive, habitat specific genome reference database. Community profiling based on this genome reference database revealed a diverse prokaryotic community comprising 61 bacterial and 18 archaeal phyla, and resulted in an improved taxonomic resolution at lower ranks down to genus level. We found that the vast majority (61 out of 73) of abundant genera (>1% average) represented unnamed and novel lineages, and that all genera could be clearly separated into brackish and marine ecotypes with inferred habitat specific functions. Applying supervised machine learning and metabolic reconstruction, we identified several microbial indicator taxa responding directly or indirectly to elevated nitrate and total phosphorus concentrations. In conclusion, our analysis highlights the importance of improved taxonomic resolution, sheds light on the role of previously uncharacterised lineages in estuarine nutrient cycling, and identifies microbial indicators for nutrient levels crucial in estuary health assessments.
- 8 January 2024
- Published by Oxford University Press (OUP) in ISME Communications
- Vol. 4 (1) , ycae067
Open AccessMachine learning and metagenomics identifies uncharacterized taxa inferred to drive biogeochemical cycles in a subtropical hypereutrophic estuary
Anthropogenic influences have drastically increased nutrient concentrations in many estuaries globally, and microbial communities have adapted to the resulting hypereutrophic ecosystems. However, our knowledge of the dominant microbial taxa and their potential functions in these ecosystems has remained sparse. Here, we study prokaryotic community dynamics in a temporal–spatial dataset, from a subtropical hypereutrophic estuary. Screening 54 water samples across brackish to marine sites revealed that nutrient concentrations and salinity best explained spatial community variations, whereas temperature and dissolved oxygen likely drive seasonal shifts. By combining short and long read sequencing data, we recovered 2,459 metagenome-assembled genomes, proposed new taxon names for previously uncharacterised lineages, and created an extensive, habitat specific genome reference database. Community profiling based on this genome reference database revealed a diverse prokaryotic community comprising 61 bacterial and 18 archaeal phyla, and resulted in an improved taxonomic resolution at lower ranks down to genus level. We found that the vast majority (61 out of 73) of abundant genera (>1% average) represented unnamed and novel lineages, and that all genera could be clearly separated into brackish and marine ecotypes with inferred habitat specific functions. Applying supervised machine learning and metabolic reconstruction, we identified several microbial indicator taxa responding directly or indirectly to elevated nitrate and total phosphorus concentrations. In conclusion, our analysis highlights the importance of improved taxonomic resolution, sheds light on the role of previously uncharacterised lineages in estuarine nutrient cycling, and identifies microbial indicators for nutrient levels crucial in estuary health assessments.