Yunxin Xu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yunxin Xu has a total of 45 co-authors. Here are the top co-authors by number of shared publications:
- Haipeng Gong
MOE Key Laboratory of Bioinformatics, School of Life Sciences Tsinghua University Beijing China
9 shared publications - Di Liu
Department of Biomedical Informatics, Columbia University Irving Medical Center, New York, USA
4 shared publications - Shize Yu
School of Software Engineering, Huazhong University of Science and Technology, Wuhan 430074, China
3 shared publications - Christopher A. Voigt
Massachusetts Institute of Technology
2 shared publications - David R. Liu
Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, USA
2 shared publications Gengjiang Liu
School of Pharmaceutical Sciences, Tsinghua University, Beijing, China
2 shared publications
Latest Publications
- Open Access
- 29 September 2026
- Published by Springer Nature in Research Square
From single-sequence structure prediction to protein fitness landscape through a composable, epistasis-aware mutation atlas
Mapping the multi-mutant fitness landscape is vital to protein engineering, but is challenging due to the vast combinatorial sequence space awaiting exploration. A central difficulty lies in the accurate and efficient modeling of non-additive epistatic effects among individual mutations, which partially arise from the physical inter-residue interactions prescribed by the protein structure. Existing fitness predictors usually perform well on single mutants but become less powerful for higher-order mutants, due to the lack of explicitly considering the relationship between sequence, structure and function of the target protein. Here, we present an end-to-end framework named Cerebra-Epistasis, which couples a single-sequence structure predictor that explicitly endows the structure awareness beyond the conventional sequence-fitness mapping with a downstream fitness prediction network that deliberately models the non-linear epistatic effects beyond the traditional additive terms. When evaluated across diverse assays, Cerebra-Epistasis outperforms the other state-of-the-art baselines in multi-mutant fitness prediction, with enhanced advantage over increasing mutation orders. Moreover, our special design on the epistasis modeling allows reliable extrapolation from low-order mutant data to unseen higher-order combinations, enabling one-shot inference of the overall mutation atlas from the starting sequence, a benefit that supposedly introduces three orders of magnitude acceleration in the landscape-scale prediction.
- 29 September 2026
- Published by Springer Nature in Research Square
Open AccessFrom single-sequence structure prediction to protein fitness landscape through a composable, epistasis-aware mutation atlas
Mapping the multi-mutant fitness landscape is vital to protein engineering, but is challenging due to the vast combinatorial sequence space awaiting exploration. A central difficulty lies in the accurate and efficient modeling of non-additive epistatic effects among individual mutations, which partially arise from the physical inter-residue interactions prescribed by the protein structure. Existing fitness predictors usually perform well on single mutants but become less powerful for higher-order mutants, due to the lack of explicitly considering the relationship between sequence, structure and function of the target protein. Here, we present an end-to-end framework named Cerebra-Epistasis, which couples a single-sequence structure predictor that explicitly endows the structure awareness beyond the conventional sequence-fitness mapping with a downstream fitness prediction network that deliberately models the non-linear epistatic effects beyond the traditional additive terms. When evaluated across diverse assays, Cerebra-Epistasis outperforms the other state-of-the-art baselines in multi-mutant fitness prediction, with enhanced advantage over increasing mutation orders. Moreover, our special design on the epistasis modeling allows reliable extrapolation from low-order mutant data to unseen higher-order combinations, enabling one-shot inference of the overall mutation atlas from the starting sequence, a benefit that supposedly introduces three orders of magnitude acceleration in the landscape-scale prediction.
- Open Access
From single-sequence structure prediction to protein fitness landscape through a composable, epistasis-aware mutation atlas
Mapping the multi-mutant fitness landscape is vital to protein engineering, but is challenging due to the vast combinatorial sequence space awaiting exploration. A central difficulty lies in the accurate and efficient modeling of non-additive epistatic effects among individual mutations, which partially arise from the physical inter-residue interactions prescribed by the protein structure. Existing fitness predictors usually perform well on single mutants but become less powerful for higher-order mutants, due to the lack of explicitly considering the relationship between sequence, structure and function of the target protein. Here, we present an end-to-end framework named Cerebra-Epistasis, which couples a single-sequence structure predictor that explicitly endows the structure awareness beyond the conventional sequence-fitness mapping with a downstream fitness prediction network that deliberately models the non-linear epistatic effects beyond the traditional additive terms. When evaluated across diverse assays, Cerebra-Epistasis outperforms the other state-of-the-art baselines in multi-mutant fitness prediction, with enhanced advantage over increasing mutation orders. Moreover, our special design on the epistasis modeling allows reliable extrapolation from low-order mutant data to unseen higher-order combinations, enabling one-shot inference of the overall mutation atlas from the starting sequence, a benefit that supposedly introduces three orders of magnitude acceleration in the landscape-scale prediction.
Open AccessFrom single-sequence structure prediction to protein fitness landscape through a composable, epistasis-aware mutation atlas
Mapping the multi-mutant fitness landscape is vital to protein engineering, but is challenging due to the vast combinatorial sequence space awaiting exploration. A central difficulty lies in the accurate and efficient modeling of non-additive epistatic effects among individual mutations, which partially arise from the physical inter-residue interactions prescribed by the protein structure. Existing fitness predictors usually perform well on single mutants but become less powerful for higher-order mutants, due to the lack of explicitly considering the relationship between sequence, structure and function of the target protein. Here, we present an end-to-end framework named Cerebra-Epistasis, which couples a single-sequence structure predictor that explicitly endows the structure awareness beyond the conventional sequence-fitness mapping with a downstream fitness prediction network that deliberately models the non-linear epistatic effects beyond the traditional additive terms. When evaluated across diverse assays, Cerebra-Epistasis outperforms the other state-of-the-art baselines in multi-mutant fitness prediction, with enhanced advantage over increasing mutation orders. Moreover, our special design on the epistasis modeling allows reliable extrapolation from low-order mutant data to unseen higher-order combinations, enabling one-shot inference of the overall mutation atlas from the starting sequence, a benefit that supposedly introduces three orders of magnitude acceleration in the landscape-scale prediction.
- Open Access
MutCleaner: Cleaning and Standardizing Biological Mutation Datasets for Variant Effect Prediction
Summary: Protein mutation datasets are widely used in variant effect prediction and protein engineering, but datasets from different sources often lack consistent conventions for mutation representation, sequence representation, data organization, and experimental labels, making these resources difficult to integrate directly and limiting their use in downstream analyses and modeling. MutCleaner is an extensible Python framework that cleans, validates, and standardizes protein- and codon-level mutation datasets through composable cleaning pipelines, unified sequence and mutation data structures, and dataset-specific cleaners. It provides standardized resources covering 16 protein and codon mutation datasets with more than 12.14 million mutation records Availability and implementation: MutCleaner is an open-source Python package released under the Apache License 2.0. The source code is available on GitHub, the package is distributed through PyPI, and the documentation is available online.
Open AccessMutCleaner: Cleaning and Standardizing Biological Mutation Datasets for Variant Effect Prediction
Summary: Protein mutation datasets are widely used in variant effect prediction and protein engineering, but datasets from different sources often lack consistent conventions for mutation representation, sequence representation, data organization, and experimental labels, making these resources difficult to integrate directly and limiting their use in downstream analyses and modeling. MutCleaner is an extensible Python framework that cleans, validates, and standardizes protein- and codon-level mutation datasets through composable cleaning pipelines, unified sequence and mutation data structures, and dataset-specific cleaners. It provides standardized resources covering 16 protein and codon mutation datasets with more than 12.14 million mutation records Availability and implementation: MutCleaner is an open-source Python package released under the Apache License 2.0. The source code is available on GitHub, the package is distributed through PyPI, and the documentation is available online.
- Open Access
- 17 March 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Calcium-mediated calreticulin-IRE1α interaction drives dynamic fluctuation of IRE1α activity under chronic endoplasmic reticulum stress
The unfolded protein response (UPR) triggered by endoplasmic reticulum (ER) stress can be both pro-survival or pro-apoptotic, depending on the duration and intensity of the stress. ER stress under (patho)physiological conditions can last for long time, yet the dynamic regulation of the UPR under prolong ER stress is largely unknown. Here, we characterized the UPR dynamics during pharmacologically induced long-term ER stress and revealed an “up-down-up” fluctuation pattern of the IRE1α signal in various types of cells. A fluctuation of the calreticulin-IRE1α interaction intensity orchestrates dynamic regulation of IRE1α activity, which negatively correlates with the intensity of the interaction between IRE1α and BIP, a known suppressor of IRE1α. The calreticulin-IRE1α interaction is negatively affected by Ca2+ concentration, which showed “down-up-down” pattern in the ER lumen over time. Furthermore, a circadian rhythmic fluctuation of calreticulin-IRE1α interaction intensity is observed in mouse liver, accompanied by oscillation of IRE1α phosphorylation level at regular physiological conditions. Our study suggests a calcium-mediated, calreticulin-driven IRE1α activity fluctuation, representing an intermediate status that the cell adopts to cope with chronic ER stress that may exist under both pathophysiological and physiological conditions.
- 17 March 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Open AccessCalcium-mediated calreticulin-IRE1α interaction drives dynamic fluctuation of IRE1α activity under chronic endoplasmic reticulum stress
The unfolded protein response (UPR) triggered by endoplasmic reticulum (ER) stress can be both pro-survival or pro-apoptotic, depending on the duration and intensity of the stress. ER stress under (patho)physiological conditions can last for long time, yet the dynamic regulation of the UPR under prolong ER stress is largely unknown. Here, we characterized the UPR dynamics during pharmacologically induced long-term ER stress and revealed an “up-down-up” fluctuation pattern of the IRE1α signal in various types of cells. A fluctuation of the calreticulin-IRE1α interaction intensity orchestrates dynamic regulation of IRE1α activity, which negatively correlates with the intensity of the interaction between IRE1α and BIP, a known suppressor of IRE1α. The calreticulin-IRE1α interaction is negatively affected by Ca2+ concentration, which showed “down-up-down” pattern in the ER lumen over time. Furthermore, a circadian rhythmic fluctuation of calreticulin-IRE1α interaction intensity is observed in mouse liver, accompanied by oscillation of IRE1α phosphorylation level at regular physiological conditions. Our study suggests a calcium-mediated, calreticulin-driven IRE1α activity fluctuation, representing an intermediate status that the cell adopts to cope with chronic ER stress that may exist under both pathophysiological and physiological conditions.
- 11 November 2024
- Published by Springer Nature in Nature Methods
- Vol. 22 (1) , 102-112
EvoAI enables extreme compression and reconstruction of the protein sequence space
Designing proteins with improved functions requires a deep understanding of how sequence and function are related, a vast space that is hard to explore. The ability to efficiently compress this space by identifying functionally important features is extremely valuable. Here we establish a method called EvoScan to comprehensively segment and scan the high-fitness sequence space to obtain anchor points that capture its essential features, especially in high dimensions. Our approach is compatible with any biomolecular function that can be coupled to a transcriptional output. We then develop deep learning and large language models to accurately reconstruct the space from these anchors, allowing computational prediction of novel, highly fit sequences without prior homology-derived or structural information. We apply this hybrid experimental–computational method, which we call EvoAI, to a repressor protein and find that only 82 anchors are sufficient to compress the high-fitness sequence space with a compression ratio of 1048. The extreme compressibility of the space informs both applied biomolecular design and understanding of natural evolution.
- 11 November 2024
- Published by Springer Nature in Nature Methods
- Vol. 22 (1) , 102-112
EvoAI enables extreme compression and reconstruction of the protein sequence space
Designing proteins with improved functions requires a deep understanding of how sequence and function are related, a vast space that is hard to explore. The ability to efficiently compress this space by identifying functionally important features is extremely valuable. Here we establish a method called EvoScan to comprehensively segment and scan the high-fitness sequence space to obtain anchor points that capture its essential features, especially in high dimensions. Our approach is compatible with any biomolecular function that can be coupled to a transcriptional output. We then develop deep learning and large language models to accurately reconstruct the space from these anchors, allowing computational prediction of novel, highly fit sequences without prior homology-derived or structural information. We apply this hybrid experimental–computational method, which we call EvoAI, to a repressor protein and find that only 82 anchors are sufficient to compress the high-fitness sequence space with a compression ratio of 1048. The extreme compressibility of the space informs both applied biomolecular design and understanding of natural evolution.
- 25 October 2024
- Published by Springer Nature in Nature Computational Science
- Vol. 4 (11) , 840-850
Improving the prediction of protein stability changes upon mutations by geometric learning and a pre-training strategy
Accurate prediction of protein mutation effects is of great importance in protein engineering and design. Here we propose GeoStab-suite, a suite of three geometric learning-based models—GeoFitness, GeoDDG and GeoDTm—for the prediction of fitness score, ΔΔG and ΔTm of a protein upon mutations, respectively. GeoFitness engages a specialized loss function to allow supervised training of a unified model using the large amount of multi-labeled fitness data in the deep mutational scanning database. To further improve the downstream tasks of ΔΔG and ΔTm prediction, the encoder of GeoFitness is reutilized as a pre-trained module in GeoDDG and GeoDTm to overcome the challenge of lacking sufficient labeled data. This pre-training strategy, in combination with data expansion, markedly improves model performance and generalizability. In the benchmark test, GeoDDG and GeoDTm outperform the other state-of-the-art methods by at least 30% and 70%, respectively, in terms of the Spearman correlation coefficient.
- 25 October 2024
- Published by Springer Nature in Nature Computational Science
- Vol. 4 (11) , 840-850
Improving the prediction of protein stability changes upon mutations by geometric learning and a pre-training strategy
Accurate prediction of protein mutation effects is of great importance in protein engineering and design. Here we propose GeoStab-suite, a suite of three geometric learning-based models—GeoFitness, GeoDDG and GeoDTm—for the prediction of fitness score, ΔΔG and ΔTm of a protein upon mutations, respectively. GeoFitness engages a specialized loss function to allow supervised training of a unified model using the large amount of multi-labeled fitness data in the deep mutational scanning database. To further improve the downstream tasks of ΔΔG and ΔTm prediction, the encoder of GeoFitness is reutilized as a pre-trained module in GeoDDG and GeoDTm to overcome the challenge of lacking sufficient labeled data. This pre-training strategy, in combination with data expansion, markedly improves model performance and generalizability. In the benchmark test, GeoDDG and GeoDTm outperform the other state-of-the-art methods by at least 30% and 70%, respectively, in terms of the Spearman correlation coefficient.
- Open Access
- 27 August 2024
- Published by Springer Nature in Nature Communications
- Vol. 15 (1) , 1-17
An end-to-end framework for the prediction of protein structure and fitness from single sequence
Significant research progress has been made in the field of protein structure and fitness prediction. Particularly, single-sequence-based structure prediction methods like ESMFold and OmegaFold achieve a balance between inference speed and prediction accuracy, showing promise for many downstream prediction tasks. Here, we propose SPIRED, a single-sequence-based structure prediction model that exhibits comparable performance to the state-of-the-art methods but with approximately 5-fold acceleration in inference and at least one order of magnitude reduction in training consumption. By integrating SPIRED with downstream neural networks, we compose an end-to-end framework named SPIRED-Fitness for the rapid prediction of both protein structure and fitness from single sequence with satisfactory accuracy. Moreover, SPIRED-Stab, the derivative of SPIRED-Fitness, achieves state-of-the-art performance in predicting the mutational effects on protein stability.
- 27 August 2024
- Published by Springer Nature in Nature Communications
- Vol. 15 (1) , 1-17
Open AccessAn end-to-end framework for the prediction of protein structure and fitness from single sequence
Significant research progress has been made in the field of protein structure and fitness prediction. Particularly, single-sequence-based structure prediction methods like ESMFold and OmegaFold achieve a balance between inference speed and prediction accuracy, showing promise for many downstream prediction tasks. Here, we propose SPIRED, a single-sequence-based structure prediction model that exhibits comparable performance to the state-of-the-art methods but with approximately 5-fold acceleration in inference and at least one order of magnitude reduction in training consumption. By integrating SPIRED with downstream neural networks, we compose an end-to-end framework named SPIRED-Fitness for the rapid prediction of both protein structure and fitness from single sequence with satisfactory accuracy. Moreover, SPIRED-Stab, the derivative of SPIRED-Fitness, achieves state-of-the-art performance in predicting the mutational effects on protein stability.
- Open Access
- 7 March 2024
- Published by Springer Nature in Research Square
SPIRED-Fitness: an end-to-end framework for the prediction of protein structure and fitness from single sequence
Significant research progress has been made in the field of protein structure and fitness prediction. Particularly, single-sequence-based structure prediction methods like ESMFold and OmegaFold achieve a balance between inference speed and prediction accuracy, showing promise for many downstream prediction tasks. Here, we propose SPIRED, a novel single-sequence-based structure prediction model that exhibits comparable performance to the state-of-the-art methods but with approximately 5-fold acceleration in inference and at least one order of magnitude reduction in training consumption. By integrating SPIRED with downstream neural networks, we compose an end-toend framework named SPIRED-Fitness for the rapid prediction of both protein structure and fitness from single sequence. SPIRED-Fitness and its derivative SPIRED-Stab achieve state-of-the-art performance in predicting the mutational effects on protein fitness and stability metrics, respectively.
- 7 March 2024
- Published by Springer Nature in Research Square
Open AccessSPIRED-Fitness: an end-to-end framework for the prediction of protein structure and fitness from single sequence
Significant research progress has been made in the field of protein structure and fitness prediction. Particularly, single-sequence-based structure prediction methods like ESMFold and OmegaFold achieve a balance between inference speed and prediction accuracy, showing promise for many downstream prediction tasks. Here, we propose SPIRED, a novel single-sequence-based structure prediction model that exhibits comparable performance to the state-of-the-art methods but with approximately 5-fold acceleration in inference and at least one order of magnitude reduction in training consumption. By integrating SPIRED with downstream neural networks, we compose an end-toend framework named SPIRED-Fitness for the rapid prediction of both protein structure and fitness from single sequence. SPIRED-Fitness and its derivative SPIRED-Stab achieve state-of-the-art performance in predicting the mutational effects on protein fitness and stability metrics, respectively.
- Open Access
- 23 February 2024
- Published by Springer Nature in Research Square
EvoAI enables extreme compression and reconstruction of the protein sequence space
Designing proteins with improved functions requires a deep understanding of how sequence and function are related, a vast space that is hard to explore. The ability to efficiently compress this space by identifying functionally important features is extremely valuable. Here, we first establish a method called EvoScan to comprehensively segment and scan the high-fitness sequence space to obtain anchor points that capture its essential features, especially in high dimensions. Our approach is compatible with any biomolecular function that can be coupled to a transcriptional output. We then develop deep learning and large language models to accurately reconstruct the space from these anchors, allowing computational prediction of novel, highly fit sequences without prior homology-derived or structural information. We apply this hybrid experimental-computational method, which we call EvoAI, to a repressor protein and find that only 82 anchors are sufficient to compress the high-fitness sequence space with a compression ratio of 1048. The extreme compressibility of the space informs both applied biomolecular design and understanding of natural evolution.
- 23 February 2024
- Published by Springer Nature in Research Square
Open AccessEvoAI enables extreme compression and reconstruction of the protein sequence space
Designing proteins with improved functions requires a deep understanding of how sequence and function are related, a vast space that is hard to explore. The ability to efficiently compress this space by identifying functionally important features is extremely valuable. Here, we first establish a method called EvoScan to comprehensively segment and scan the high-fitness sequence space to obtain anchor points that capture its essential features, especially in high dimensions. Our approach is compatible with any biomolecular function that can be coupled to a transcriptional output. We then develop deep learning and large language models to accurately reconstruct the space from these anchors, allowing computational prediction of novel, highly fit sequences without prior homology-derived or structural information. We apply this hybrid experimental-computational method, which we call EvoAI, to a repressor protein and find that only 82 anchors are sufficient to compress the high-fitness sequence space with a compression ratio of 1048. The extreme compressibility of the space informs both applied biomolecular design and understanding of natural evolution.
- Open Access
SPIRED-Fitness: an end-to-end framework for the prediction of protein structure and fitness from single sequence
A bstract: Significant research progress has been made in the field of protein structure and fitness prediction. Particularly, single-sequence-based structure prediction methods like ESMFold and OmegaFold achieve a balance between inference speed and prediction accuracy, showing promise for many downstream prediction tasks. Here, we propose SPIRED, a novel single-sequence-based structure prediction model that exhibits comparable performance to the state-of-the-art methods but with approximately 5-fold acceleration in inference and at least one order of magnitude reduction in training consumption. By integrating SPIRED with downstream neural networks, we compose an end-to-end framework named SPIRED-Fitness for the rapid prediction of both protein structure and fitness from single sequence. SPIRED-Fitness and its derivative SPIRED-Stab achieve state-of-the-art performance in predicting the mutational effects on protein fitness and stability metrics, respectively.
Open AccessSPIRED-Fitness: an end-to-end framework for the prediction of protein structure and fitness from single sequence
A bstract: Significant research progress has been made in the field of protein structure and fitness prediction. Particularly, single-sequence-based structure prediction methods like ESMFold and OmegaFold achieve a balance between inference speed and prediction accuracy, showing promise for many downstream prediction tasks. Here, we propose SPIRED, a novel single-sequence-based structure prediction model that exhibits comparable performance to the state-of-the-art methods but with approximately 5-fold acceleration in inference and at least one order of magnitude reduction in training consumption. By integrating SPIRED with downstream neural networks, we compose an end-to-end framework named SPIRED-Fitness for the rapid prediction of both protein structure and fitness from single sequence. SPIRED-Fitness and its derivative SPIRED-Stab achieve state-of-the-art performance in predicting the mutational effects on protein fitness and stability metrics, respectively.