Itoshi Nikaido
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Itoshi Nikaido has a total of 430 co-authors. Here are the top co-authors by number of shared publications:
- Tetsutaro Hayashi
Laboratory for Bioinformatics Research, RIKEN Center for Biosystems Dynamics Research, Wako, Japan
19 shared publications - Yohei Sasagawa
Laboratory for Bioinformatics Research, RIKEN Center for Biosystems Dynamics Research, Wako, Japan
15 shared publications - Koki Tsuyuzaki
Department of Artificial Intelligence Medicine, Graduate School of Medicine, Chiba University, Japan
11 shared publications Mika Yoshimura
Omics AI Research Team, Advanced General Intelligence in Science Program (AGIS), TRIP Headquarters, RIKEN, Kobe, Hyogo 650-0047, Japan
9 shared publicationsMana Umeda
Laboratory for Bioinformatics Research, RIKEN Center for Biosystems Dynamics Research, Kobe, Hyogo 650-0047, Japan
8 shared publicationsYasushi Okazaki
Saitama Medical University
8 shared publications
Latest Publications
- Open Access
- 16 October 2026
- Published by Springer Nature in BMC Bioinformatics
GenPerturb: sequence-grounded interpretation of perturbation transcriptomes using pretrained genomic models
Background: Perturb-seq captures transcriptional responses to thousands of genetic and chemical perturbations, but does not directly resolve the cis-regulatory elements or transcription factor motifs underlying those responses. Existing approaches rely on indirect post hoc analyses or external epigenomic annotations, making it difficult to connect gene-level responses to specific regulatory elements. Results: We present GenPerturb, a model-based interpretation workflow that fits condition-specific linear readouts of pretrained sequence embeddings to observed Perturb-seq pseudobulk expression. Because control and perturbation readouts operate on one gene-centred sequence representation, control-subtracted attribution ranks candidate regions and motif instances as computational hypotheses associated with each fitted contrast. The fitted readouts recover much of the observed expression variation between genes, whereas recovery of variation across perturbation conditions within each gene is more modest. On the single available erythroid Perturb-multiome benchmark, region rankings showed agreement with accessibility labels comparable to annotation-based scores, with slightly closer agreement in the 1–10 and 10–100 kb strata, while the perturbed transcription factor’s motif family was recovered at comparable frequencies from attribution-derived and chromatin-derived regions. Selected lineage and compound examples illustrate motif associations that can differ from those of the corresponding transcription factor mRNA. Conclusions: GenPerturb transfers regulatory priors learned during external pretraining to observed RNA-only Perturb-seq profiles, producing ranked, locus-anchored cis-regulatory hypotheses without a matched chromatin assay in the target experiment. Scoped to observed responses rather than prediction of unseen perturbations or cell types, the workflow offers computational candidates for experimental assessment of enhancer and transcription-factor activity.
- 16 October 2026
- Published by Springer Nature in BMC Bioinformatics
Open AccessGenPerturb: sequence-grounded interpretation of perturbation transcriptomes using pretrained genomic models
Background: Perturb-seq captures transcriptional responses to thousands of genetic and chemical perturbations, but does not directly resolve the cis-regulatory elements or transcription factor motifs underlying those responses. Existing approaches rely on indirect post hoc analyses or external epigenomic annotations, making it difficult to connect gene-level responses to specific regulatory elements. Results: We present GenPerturb, a model-based interpretation workflow that fits condition-specific linear readouts of pretrained sequence embeddings to observed Perturb-seq pseudobulk expression. Because control and perturbation readouts operate on one gene-centred sequence representation, control-subtracted attribution ranks candidate regions and motif instances as computational hypotheses associated with each fitted contrast. The fitted readouts recover much of the observed expression variation between genes, whereas recovery of variation across perturbation conditions within each gene is more modest. On the single available erythroid Perturb-multiome benchmark, region rankings showed agreement with accessibility labels comparable to annotation-based scores, with slightly closer agreement in the 1–10 and 10–100 kb strata, while the perturbed transcription factor’s motif family was recovered at comparable frequencies from attribution-derived and chromatin-derived regions. Selected lineage and compound examples illustrate motif associations that can differ from those of the corresponding transcription factor mRNA. Conclusions: GenPerturb transfers regulatory priors learned during external pretraining to observed RNA-only Perturb-seq profiles, producing ranked, locus-anchored cis-regulatory hypotheses without a matched chromatin assay in the target experiment. Scoped to observed responses rather than prediction of unseen perturbations or cell types, the workflow offers computational candidates for experimental assessment of enhancer and transcription-factor activity.
- Open Access
GenPerturb: sequence-grounded interpretation of perturbation transcriptomes using pretrained genomic models
Background: Perturb-seq captures transcriptional responses to thousands of genetic and chemical perturbations, but does not directly resolve the cis-regulatory elements or transcription factor motifs underlying those responses. Existing approaches rely on indirect post hoc analyses or external epigenomic annotations, making it difficult to connect gene-level responses to specific regulatory element Results: We present GenPerturb, a framework that leverages pretrained sequence-to-expression models to link perturbation-induced expression changes to candidate cis-regulatory elements. By contrasting perturbation and control states, GenPerturb prioritizes regulatory regions and transcription factor motifs associated with each perturbation. The model recapitulates perturbation-dependent gene expression patterns and enables sequence-level interpretation without requiring matched chromatin data. Across multiple perturbation types, GenPerturb identifies biologically meaningful regulatory programs, including lineage-specific and signaling-associated motif activities, even when corresponding transcription factor expression changes are limited. Conclusions: GenPerturb converts gene-level expression responses from Perturb-seq into perturbation-specific, sequence-grounded cis-regulatory hypotheses. By prioritizing candidate regulatory elements and transcription factor motifs responsive to each perturbation without requiring matched chromatin data, GenPerturb enables mechanistic interpretation of transcriptional regulation and guides downstream experimental validation.
Open AccessGenPerturb: sequence-grounded interpretation of perturbation transcriptomes using pretrained genomic models
Background: Perturb-seq captures transcriptional responses to thousands of genetic and chemical perturbations, but does not directly resolve the cis-regulatory elements or transcription factor motifs underlying those responses. Existing approaches rely on indirect post hoc analyses or external epigenomic annotations, making it difficult to connect gene-level responses to specific regulatory element Results: We present GenPerturb, a framework that leverages pretrained sequence-to-expression models to link perturbation-induced expression changes to candidate cis-regulatory elements. By contrasting perturbation and control states, GenPerturb prioritizes regulatory regions and transcription factor motifs associated with each perturbation. The model recapitulates perturbation-dependent gene expression patterns and enables sequence-level interpretation without requiring matched chromatin data. Across multiple perturbation types, GenPerturb identifies biologically meaningful regulatory programs, including lineage-specific and signaling-associated motif activities, even when corresponding transcription factor expression changes are limited. Conclusions: GenPerturb converts gene-level expression responses from Perturb-seq into perturbation-specific, sequence-grounded cis-regulatory hypotheses. By prioritizing candidate regulatory elements and transcription factor motifs responsive to each perturbation without requiring matched chromatin data, GenPerturb enables mechanistic interpretation of transcriptional regulation and guides downstream experimental validation.
- Open Access
Cycling persister clones with elevated NR2F1-mediated cholesterol biosynthesis cause chemotherapy resistance
While a subpopulation termed cycling persisters (CPs) that is characterized by sustained proliferation, even under chemotherapy drug exposure, contributes more directly to tumor relapse, the molecular basis for the emergence of CPs has been unclear. Here, we used the human tongue cancer organoid (TCO) library to continuously track the in vitro fate of individual cancer cell clones during and after chemotherapy exposure using time-lapse imaging. Among the heterogeneous clones, we identified CPs that formed larger clusters than the others, which barely grew and remained small (non-CPs). Using differences in cell cluster size as an indicator, thousands of CP and non-CP clones were directly sampled from 3D matrix organoid cultures and were analyzed. Notably, tumor-intrinsic interferon (IFN) signaling and hypoxic pathways were inactivated, whereas the NR2F1-mediated cholesterol biosynthesis pathway was distinctly activated in CPs compared to non-CPs. Indeed, inhibiting cholesterol biosynthesis with simvastatin significantly suppressed the appearance of CP clones, showing that elevated cholesterol biosynthesis is essential for the emergence of CPs. These findings suggest that clonal-level variations in the intensity of these signaling pathways determine the fate of individual tumor cells exposed to chemotherapeutic agents, which may provide insights into cancer relapse mechanisms and identify potential molecular targets of CPs.
Open AccessCycling persister clones with elevated NR2F1-mediated cholesterol biosynthesis cause chemotherapy resistance
While a subpopulation termed cycling persisters (CPs) that is characterized by sustained proliferation, even under chemotherapy drug exposure, contributes more directly to tumor relapse, the molecular basis for the emergence of CPs has been unclear. Here, we used the human tongue cancer organoid (TCO) library to continuously track the in vitro fate of individual cancer cell clones during and after chemotherapy exposure using time-lapse imaging. Among the heterogeneous clones, we identified CPs that formed larger clusters than the others, which barely grew and remained small (non-CPs). Using differences in cell cluster size as an indicator, thousands of CP and non-CP clones were directly sampled from 3D matrix organoid cultures and were analyzed. Notably, tumor-intrinsic interferon (IFN) signaling and hypoxic pathways were inactivated, whereas the NR2F1-mediated cholesterol biosynthesis pathway was distinctly activated in CPs compared to non-CPs. Indeed, inhibiting cholesterol biosynthesis with simvastatin significantly suppressed the appearance of CP clones, showing that elevated cholesterol biosynthesis is essential for the emergence of CPs. These findings suggest that clonal-level variations in the intensity of these signaling pathways determine the fate of individual tumor cells exposed to chemotherapeutic agents, which may provide insights into cancer relapse mechanisms and identify potential molecular targets of CPs.
- Open Access
- 15 December 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 10902
scRepli-RamDA-seq: a multi-omics technology enabling the analysis of gene expression dynamics during S-phase
Single-cell sequencing has advanced our understanding of cell-type diversity and heterogeneity. However, existing single-cell multi-omics methods lack the ability to monitor gene expression dynamics during S-phase progression. Here, we introduce single-cell (sc)Repli-RamDA-seq (scRR-seq), a multi-omics method that enables high-resolution DNA replication profiling and full-length total RNA sequencing from the same single cell in a haplotype-specific manner. scRR-seq generates DNA replication and RNA sequencing data comparable to individually obtained scRepli-seq and scRamDA-seq data, respectively. Unlike other scDNA/RNA-seq methods, scRR-seq allows one to tell the S-phase stage of a given cell based on the percentage of the replicated genome derived from scRepli-seq data. This facilitates analysis of gene expression dynamics during S-phase progression, enabling the identification of S-phase progression markers. scRR-seq also detects copy-number variation in non-S-phase cells and outperforms other scDNA/RNA-seq methods in various measures. Taken together, scRR-seq is a robust single-cell multi-omics method with promising potential for comprehensive genome/transcriptome analysis.
- 15 December 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 10902
Open AccessscRepli-RamDA-seq: a multi-omics technology enabling the analysis of gene expression dynamics during S-phase
Single-cell sequencing has advanced our understanding of cell-type diversity and heterogeneity. However, existing single-cell multi-omics methods lack the ability to monitor gene expression dynamics during S-phase progression. Here, we introduce single-cell (sc)Repli-RamDA-seq (scRR-seq), a multi-omics method that enables high-resolution DNA replication profiling and full-length total RNA sequencing from the same single cell in a haplotype-specific manner. scRR-seq generates DNA replication and RNA sequencing data comparable to individually obtained scRepli-seq and scRamDA-seq data, respectively. Unlike other scDNA/RNA-seq methods, scRR-seq allows one to tell the S-phase stage of a given cell based on the percentage of the replicated genome derived from scRepli-seq data. This facilitates analysis of gene expression dynamics during S-phase progression, enabling the identification of S-phase progression markers. scRR-seq also detects copy-number variation in non-S-phase cells and outperforms other scDNA/RNA-seq methods in various measures. Taken together, scRR-seq is a robust single-cell multi-omics method with promising potential for comprehensive genome/transcriptome analysis.
- Open Access
- 25 September 2025
- Published by Springer Nature in BMC Genomics
- Vol. 26 (1) , 1-15
MitoDelta: identifying mitochondrial DNA deletions at cell-type resolution from single-cell RNA sequencing data
Background: Deletion variants in mitochondrial DNA (mtDNA) are associated with various diseases, such as mitochondrial disorders and neurodegenerative diseases. Traditionally, mtDNA deletions have been studied using bulk DNA sequencing, but bulk methods average signals across cells, thereby masking the cell-type-specific mutational landscapes. Resolving mtDNA deletions at single-cell resolution is beneficial for understanding how these mutations affect distinct cell populations. To date, no specialized method exists for detecting cell-type-specific mtDNA deletions from single-cell RNA sequencing data. Notably, mtDNA possesses unique molecular features: a high copy number, stable transcription, and compact structure of the mitochondrial genome. This results in a relatively high abundance of mtDNA-derived reads even in single-cell RNA sequencing data, suggesting the possibility of detecting mtDNA deletion variants directly from transcriptomic data. Results: Here, we present MitoDelta, a computational pipeline that enables the detection of mtDNA deletions at cell-type resolution solely from single-cell RNA sequencing data. MitoDelta combines a sensitive alignment strategy with robust statistical filtering based on a beta-binomial distribution model, allowing accurate identification of deletion events even from noisy single-cell transcriptomes. To capture cell-type-specific deletion patterns, MitoDelta analyzes reads pooled by annotated cell types, enabling quantification of deletion burden across distinct cellular populations. We benchmarked MitoDelta against existing mtDNA deletion detection tools and demonstrated superior overall performance. As a practical application, we applied MitoDelta to a published single-nucleus RNA sequencing dataset for Parkinson’s disease and revealed distinct mtDNA deletion burdens across neuronal subtypes. Conclusions: MitoDelta enables the transcriptome-integrated, cell-type-specific detection of mtDNA deletions from single-cell RNA sequencing data alone, offering a valuable framework for reanalyzing public datasets and studying mitochondrial genome alterations at cell-type resolution. This integrated approach enables insights into how mtDNA deletions are distributed across specific cell types and cellular states, providing new opportunities to investigate the role of mtDNA deletions in cell-type-specific disease mechanisms. The tool is available at https://github.com/NikaidoLaboratory/mitodelta.
- 25 September 2025
- Published by Springer Nature in BMC Genomics
- Vol. 26 (1) , 1-15
Open AccessMitoDelta: identifying mitochondrial DNA deletions at cell-type resolution from single-cell RNA sequencing data
Background: Deletion variants in mitochondrial DNA (mtDNA) are associated with various diseases, such as mitochondrial disorders and neurodegenerative diseases. Traditionally, mtDNA deletions have been studied using bulk DNA sequencing, but bulk methods average signals across cells, thereby masking the cell-type-specific mutational landscapes. Resolving mtDNA deletions at single-cell resolution is beneficial for understanding how these mutations affect distinct cell populations. To date, no specialized method exists for detecting cell-type-specific mtDNA deletions from single-cell RNA sequencing data. Notably, mtDNA possesses unique molecular features: a high copy number, stable transcription, and compact structure of the mitochondrial genome. This results in a relatively high abundance of mtDNA-derived reads even in single-cell RNA sequencing data, suggesting the possibility of detecting mtDNA deletion variants directly from transcriptomic data. Results: Here, we present MitoDelta, a computational pipeline that enables the detection of mtDNA deletions at cell-type resolution solely from single-cell RNA sequencing data. MitoDelta combines a sensitive alignment strategy with robust statistical filtering based on a beta-binomial distribution model, allowing accurate identification of deletion events even from noisy single-cell transcriptomes. To capture cell-type-specific deletion patterns, MitoDelta analyzes reads pooled by annotated cell types, enabling quantification of deletion burden across distinct cellular populations. We benchmarked MitoDelta against existing mtDNA deletion detection tools and demonstrated superior overall performance. As a practical application, we applied MitoDelta to a published single-nucleus RNA sequencing dataset for Parkinson’s disease and revealed distinct mtDNA deletion burdens across neuronal subtypes. Conclusions: MitoDelta enables the transcriptome-integrated, cell-type-specific detection of mtDNA deletions from single-cell RNA sequencing data alone, offering a valuable framework for reanalyzing public datasets and studying mitochondrial genome alterations at cell-type resolution. This integrated approach enables insights into how mtDNA deletions are distributed across specific cell types and cellular states, providing new opportunities to investigate the role of mtDNA deletions in cell-type-specific disease mechanisms. The tool is available at https://github.com/NikaidoLaboratory/mitodelta.
- Open Access
Elucidation of the mechanism of byproduct DNA synthesis in whole-transcript amplification by poly-A tagging and development of its specific suppression
Poly-A tagging is a whole-transcript amplification method that uses terminal transferase to add poly-A to single-stranded cDNA in a non-template-dependent manner, and is used in highly sensitive single-cell RNA-seq. However, this method has long been reported to amplify byproduct DNA derived from reverse-transcription primers, particularly under conditions of low RNA abundance, leading to experimental complexity and reduced accuracy in quantifying gene expression. In this study, we elucidated the mechanism by which byproduct DNA is amplified at the molecular level and developed a new strategy of suppressing it named Suppression Tagging (SupTag). SupTag makes use of TdT, which has extremely low extension efficiency at the 3′ recessive end, to stabilize dA/dT base pairs and inhibit poly-A extension of oligo-dT primers, thereby preventing the formation of precursor DNA for byproduct DNA. We demonstrated that SupTag can specifically suppress byproduct DNA using chemicals such as tetramethylammonium chloride (TMAC), modified nucleotides (e.g., 2F-dATP), and Super T-modified reverse-transcription primers. SupTag improves usability by streamlining purification processes and prevents reductions in sequence read output and quantitative sensitivity caused by byproduct DNA. Additionally, SupTag demonstrated the ability to suppress byproduct DNA amplification without using suppression PCR and is compatible with cDNA amplification introducing different adapter sequences at both ends. This study presents a novel suppression method based on the mechanism of amplification of byproduct DNA in the poly-A tagging method, providing a practical technology enabling high-precision transcriptomic analysis from low-abundance RNA.
Open AccessElucidation of the mechanism of byproduct DNA synthesis in whole-transcript amplification by poly-A tagging and development of its specific suppression
Poly-A tagging is a whole-transcript amplification method that uses terminal transferase to add poly-A to single-stranded cDNA in a non-template-dependent manner, and is used in highly sensitive single-cell RNA-seq. However, this method has long been reported to amplify byproduct DNA derived from reverse-transcription primers, particularly under conditions of low RNA abundance, leading to experimental complexity and reduced accuracy in quantifying gene expression. In this study, we elucidated the mechanism by which byproduct DNA is amplified at the molecular level and developed a new strategy of suppressing it named Suppression Tagging (SupTag). SupTag makes use of TdT, which has extremely low extension efficiency at the 3′ recessive end, to stabilize dA/dT base pairs and inhibit poly-A extension of oligo-dT primers, thereby preventing the formation of precursor DNA for byproduct DNA. We demonstrated that SupTag can specifically suppress byproduct DNA using chemicals such as tetramethylammonium chloride (TMAC), modified nucleotides (e.g., 2F-dATP), and Super T-modified reverse-transcription primers. SupTag improves usability by streamlining purification processes and prevents reductions in sequence read output and quantitative sensitivity caused by byproduct DNA. Additionally, SupTag demonstrated the ability to suppress byproduct DNA amplification without using suppression PCR and is compatible with cDNA amplification introducing different adapter sequences at both ends. This study presents a novel suppression method based on the mechanism of amplification of byproduct DNA in the poly-A tagging method, providing a practical technology enabling high-precision transcriptomic analysis from low-abundance RNA.
- Open Access
Genomic characterization of sub-populations in human pluripotent stem cell-derived retinal progenitor cells that drive retinal layer structure
SUMMARY: The mechanism underlying retinal spheroid layer formation was investigated using Rax::GFP-positive retinal progenitor cells from human embryonic stem cell-derived retinal organoids. Single-cell RNA sequencing revealed that well-layered spheroids transiently activated canonical WNT2B–FZD7 signaling followed by temporary expression of non-canonical WNT5A. Despite structural differences in vitro, non- layered and well-layered retinal spheroids on differentiation Day 60 successfully developed into cone and rod subtypes of photoreceptors and established synaptic connections with host bipolar cells after transplantation in a retinal degeneration rat model, resulting in light-evoked functional recovery. Additionally, part of the presumable Rax::GFP-positive retinal progenitor cells differentiated into non- retinal lineages, including ciliary marginal zone-like, retinal pigment epithelium, and spinal cord-like tissues in vitro, reflecting the developmental plasticity of RAX-positive cells. These findings suggest that canonical and non-canonical WNT signaling pathways sequentially orchestrate early retinal morphogenesis, whereas environmental factors within the host retina strongly drive the alignment and functional integration of graft photoreceptors.
Open AccessGenomic characterization of sub-populations in human pluripotent stem cell-derived retinal progenitor cells that drive retinal layer structure
SUMMARY: The mechanism underlying retinal spheroid layer formation was investigated using Rax::GFP-positive retinal progenitor cells from human embryonic stem cell-derived retinal organoids. Single-cell RNA sequencing revealed that well-layered spheroids transiently activated canonical WNT2B–FZD7 signaling followed by temporary expression of non-canonical WNT5A. Despite structural differences in vitro, non- layered and well-layered retinal spheroids on differentiation Day 60 successfully developed into cone and rod subtypes of photoreceptors and established synaptic connections with host bipolar cells after transplantation in a retinal degeneration rat model, resulting in light-evoked functional recovery. Additionally, part of the presumable Rax::GFP-positive retinal progenitor cells differentiated into non- retinal lineages, including ciliary marginal zone-like, retinal pigment epithelium, and spinal cord-like tissues in vitro, reflecting the developmental plasticity of RAX-positive cells. These findings suggest that canonical and non-canonical WNT signaling pathways sequentially orchestrate early retinal morphogenesis, whereas environmental factors within the host retina strongly drive the alignment and functional integration of graft photoreceptors.
- Open Access
MitoDelta: identifying mitochondrial DNA deletions at cell-type resolution from single-cell RNA sequencing data
Background Deletion variants in mitochondrial DNA (mtDNA) are associated with various diseases, such as mitochondrial disorders and neurodegenerative diseases. Traditionally, mtDNA deletions have been studied using bulk DNA sequencing, but bulk methods average signals across cells, thereby masking the cell-type-specific mutational landscapes. Resolving mtDNA deletions at single-cell resolution is beneficial for understanding how these mutations affect distinct cell populations. To date, no specialized method exists for detecting cell-type-specific mtDNA deletions from single-cell RNA sequencing data. Notably, mtDNA possesses unique molecular features: a high copy number, stable transcription, and compact structure of the mitochondrial genome. This results in a relatively high abundance of mtDNA-derived reads even in single-cell RNA sequencing data, suggesting the possibility of detecting mtDNA deletion variants directly from transcriptomic data. Results Here, we present MitoDelta, a computational pipeline that enables the detection of mtDNA deletions at cell-type resolution solely from single-cell RNA sequencing data. MitoDelta combines a sensitive alignment strategy with robust statistical filtering based on a beta-binomial model, allowing accurate identification of deletion events even from noisy single-cell transcriptomes. To capture cell-type-specific deletion patterns, MitoDelta analyzes reads pooled by annotated cell types, enabling quantification of deletion burden across distinct cellular populations. We benchmarked MitoDelta against existing mtDNA deletion detection tools and demonstrated superior overall performance. As a practical application, we applied MitoDelta to a published single-nucleus RNA sequencing dataset for Parkinson’s disease and revealed distinct mtDNA deletion burdens across neuronal subtypes. Conclusions MitoDelta enables the transcriptome-integrated, cell-type-specific detection of mtDNA deletions from single-cell RNA sequencing data alone, offering a valuable framework for reanalyzing public datasets and studying mitochondrial genome alterations at cell-type resolution. This integrated approach enables insights into how mtDNA deletions are distributed across specific cell types and cellular states, providing new opportunities to investigate the role of mtDNA deletions in cell-type-specific disease mechanisms. The tool is available at https://github.com/NikaidoLaboratory/mitodelta.
Open AccessMitoDelta: identifying mitochondrial DNA deletions at cell-type resolution from single-cell RNA sequencing data
Background Deletion variants in mitochondrial DNA (mtDNA) are associated with various diseases, such as mitochondrial disorders and neurodegenerative diseases. Traditionally, mtDNA deletions have been studied using bulk DNA sequencing, but bulk methods average signals across cells, thereby masking the cell-type-specific mutational landscapes. Resolving mtDNA deletions at single-cell resolution is beneficial for understanding how these mutations affect distinct cell populations. To date, no specialized method exists for detecting cell-type-specific mtDNA deletions from single-cell RNA sequencing data. Notably, mtDNA possesses unique molecular features: a high copy number, stable transcription, and compact structure of the mitochondrial genome. This results in a relatively high abundance of mtDNA-derived reads even in single-cell RNA sequencing data, suggesting the possibility of detecting mtDNA deletion variants directly from transcriptomic data. Results Here, we present MitoDelta, a computational pipeline that enables the detection of mtDNA deletions at cell-type resolution solely from single-cell RNA sequencing data. MitoDelta combines a sensitive alignment strategy with robust statistical filtering based on a beta-binomial model, allowing accurate identification of deletion events even from noisy single-cell transcriptomes. To capture cell-type-specific deletion patterns, MitoDelta analyzes reads pooled by annotated cell types, enabling quantification of deletion burden across distinct cellular populations. We benchmarked MitoDelta against existing mtDNA deletion detection tools and demonstrated superior overall performance. As a practical application, we applied MitoDelta to a published single-nucleus RNA sequencing dataset for Parkinson’s disease and revealed distinct mtDNA deletion burdens across neuronal subtypes. Conclusions MitoDelta enables the transcriptome-integrated, cell-type-specific detection of mtDNA deletions from single-cell RNA sequencing data alone, offering a valuable framework for reanalyzing public datasets and studying mitochondrial genome alterations at cell-type resolution. This integrated approach enables insights into how mtDNA deletions are distributed across specific cell types and cellular states, providing new opportunities to investigate the role of mtDNA deletions in cell-type-specific disease mechanisms. The tool is available at https://github.com/NikaidoLaboratory/mitodelta.
- Open Access
- 5 May 2025
- Published by Springer Nature in Research Square
Transfer learning using generative pretrained genomic DNA models for predicting perturbation-induced changes in gene expression
Background Transfer learning applied to genomic DNA models has the potential to improve predictive capabilities, especially when target-domain datasets and computational resources are limited. Despite its promise, the practical effectiveness of transfer learning in genomic DNA models, particularly for predicting gene expression changes due to perturbations, has not been thoroughly investigated. This study aimed to systematically evaluate the performance and utility of transfer learning approaches using genomic DNA models to accurately predict perturbation-induced gene expression. Results We benchmarked three genomic DNA models across 12 distinct datasets containing perturbation-induced gene expression data to identify optimal conditions for effective transfer learning. Notably, perturbation-induced gene expression data were not included in the pre-training of these genomic DNA models. Among these, the Enformer model consistently generated accurate embeddings, demonstrating superior clustering performance and gene signature scoring aligned closely with observed experimental data. Additionally, we identified a phenomenon termed "genomic neighbouring gene interference," wherein partially overlapping DNA sequences of adjacent genes cause correlated predictions, resulting in both beneficial and detrimental effects on predictive accuracy. Conclusions Our findings highlight the efficacy of transfer learning in genomic DNA models for predicting perturbation-induced gene expression, particularly emphasizing the Enformer model's robust performance. Understanding genomic neighbouring gene interference offers critical insights for refining predictive accuracy in genomic applications. This study provides practical guidance for researchers developing transfer learning strategies and genomic DNA models, paving the way for more accurate and resource-efficient genomic predictions.
- 5 May 2025
- Published by Springer Nature in Research Square
Open AccessTransfer learning using generative pretrained genomic DNA models for predicting perturbation-induced changes in gene expression
Background Transfer learning applied to genomic DNA models has the potential to improve predictive capabilities, especially when target-domain datasets and computational resources are limited. Despite its promise, the practical effectiveness of transfer learning in genomic DNA models, particularly for predicting gene expression changes due to perturbations, has not been thoroughly investigated. This study aimed to systematically evaluate the performance and utility of transfer learning approaches using genomic DNA models to accurately predict perturbation-induced gene expression. Results We benchmarked three genomic DNA models across 12 distinct datasets containing perturbation-induced gene expression data to identify optimal conditions for effective transfer learning. Notably, perturbation-induced gene expression data were not included in the pre-training of these genomic DNA models. Among these, the Enformer model consistently generated accurate embeddings, demonstrating superior clustering performance and gene signature scoring aligned closely with observed experimental data. Additionally, we identified a phenomenon termed "genomic neighbouring gene interference," wherein partially overlapping DNA sequences of adjacent genes cause correlated predictions, resulting in both beneficial and detrimental effects on predictive accuracy. Conclusions Our findings highlight the efficacy of transfer learning in genomic DNA models for predicting perturbation-induced gene expression, particularly emphasizing the Enformer model's robust performance. Understanding genomic neighbouring gene interference offers critical insights for refining predictive accuracy in genomic applications. This study provides practical guidance for researchers developing transfer learning strategies and genomic DNA models, paving the way for more accurate and resource-efficient genomic predictions.
- 7 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Biology and Bioinformatics
- Vol. 22 (2) , 570-580
The Effect of Multi-Task Learning on the Prediction of Neoantigen–MHC Class II Binding
Neoepitopes are significant therapeutic cancer vaccine candidates, given that tumor neoepitopes induce an immune response to eliminate cancer cells. This immune activation depends on the binding affinity between the antigen peptide and the major histocompatibility complex (MHC). The epitope–MHC binding assay is a technologically difficult, time-consuming, and expensive technique. Therefore, prediction methods for these binding affinities have been developed using computational prediction approaches. However, these predictive models are trained on datasets biased toward viral peptides and some MHC alleles and are limited in their prediction of neoepitopes. In particular, because of the wide variety of MHC class II binding formats, the performance of MHC class II prediction must be improved. Here, we propose a novel deep learning model that consists of multi-task bidirectional long short-term memory (Bi-LSTM) models. Our multi-task model can predict neoepitope–MHC class II bindings from limited training data by sharing MHC class I and II training parameters. We confirm that multi-task learning significantly enhances the prediction performances of cancer antigens. Our model achieves an area under the receiver operating characteristic curve (AUC-ROC) of 82.2%, outperforming existing state-of-the-art single allele neoantigen prediction models while maintaining its strong generalization performance. Our code and data are available at https://github.com/rikenbit/MTL4MHC2.
- 7 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Biology and Bioinformatics
- Vol. 22 (2) , 570-580
The Effect of Multi-Task Learning on the Prediction of Neoantigen–MHC Class II Binding
Neoepitopes are significant therapeutic cancer vaccine candidates, given that tumor neoepitopes induce an immune response to eliminate cancer cells. This immune activation depends on the binding affinity between the antigen peptide and the major histocompatibility complex (MHC). The epitope–MHC binding assay is a technologically difficult, time-consuming, and expensive technique. Therefore, prediction methods for these binding affinities have been developed using computational prediction approaches. However, these predictive models are trained on datasets biased toward viral peptides and some MHC alleles and are limited in their prediction of neoepitopes. In particular, because of the wide variety of MHC class II binding formats, the performance of MHC class II prediction must be improved. Here, we propose a novel deep learning model that consists of multi-task bidirectional long short-term memory (Bi-LSTM) models. Our multi-task model can predict neoepitope–MHC class II bindings from limited training data by sharing MHC class I and II training parameters. We confirm that multi-task learning significantly enhances the prediction performances of cancer antigens. Our model achieves an area under the receiver operating characteristic curve (AUC-ROC) of 82.2%, outperforming existing state-of-the-art single allele neoantigen prediction models while maintaining its strong generalization performance. Our code and data are available at https://github.com/rikenbit/MTL4MHC2.