Yuzhi Guo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yuzhi Guo has a total of 28 co-authors. Here are the top co-authors by number of shared publications:
- Hehuan Ma
The University of Texas at Arlington
7 shared publications - Junzhou Huang
The University of Texas at Arlington
7 shared publications Feng Jiang
2 shared publications- Jean Gao
The University of Texas at Arlington
2 shared publications - Saiyang Na
Department of Computer Science and Engineering, The University of Texas at Arlington, Arlington, TX, USA
2 shared publications - Weizhi An
Department of Computer Science and Engineering, University of Texas at Arlington, Arlington, TX 76019, USA
2 shared publications
Latest Publications
- 12 September 2026
- Published by Springer Nature
- p. 408-425
Histopathology Multi-modal Embedding for Pathology Composed Retrieval
- 12 September 2026
- Published by Springer Nature
- p. 408-425
Histopathology Multi-modal Embedding for Pathology Composed Retrieval
- Open Access
Compositional Image Retrieval via Instruction-Aware Contrastive Learning
Composed Image Retrieval (CIR) involves retrieving a target image based on a composed query of an image paired with text that specifies modifications or changes to the visual reference. CIR is inherently an instruction-following task, as the model needs to interpret and apply modifications to the image. In practice, due to the scarcity of annotated data in downstream tasks, Zero-Shot CIR (ZS-CIR) is desirable. While existing ZS-CIR models based on CLIP have shown promising results, their capability in interpreting and following modification instructions remains limited. Some research attempts to address this by incorporating Large Language Models (LLMs). However, these approaches still face challenges in effectively integrating multimodal information and instruction understanding. To tackle above challenges, we propose a novel embedding method utilizing an instruction-tuned Multimodal LLM (MLLM) to generate composed representation, which significantly enhance the instruction following capability for a comprehensive integration between images and instructions. Nevertheless, directly applying MLLMs introduces a new challenge since MLLMs are primarily designed for text generation rather than embedding extraction as required in CIR. To address this, we introduce a two-stage training strategy to efficiently learn a joint multimodal embedding space and further refining the ability to follow modification instructions by tuning the model in a triplet dataset similar to the CIR format. Extensive experiments on four public datasets: FashionIQ, CIRR, GeneCIS, and CIRCO demonstrates the superior performance of our model, outperforming state-of-the-art baselines by a significant margin. Codes are available at the GitHub repository.
Open AccessCompositional Image Retrieval via Instruction-Aware Contrastive Learning
Composed Image Retrieval (CIR) involves retrieving a target image based on a composed query of an image paired with text that specifies modifications or changes to the visual reference. CIR is inherently an instruction-following task, as the model needs to interpret and apply modifications to the image. In practice, due to the scarcity of annotated data in downstream tasks, Zero-Shot CIR (ZS-CIR) is desirable. While existing ZS-CIR models based on CLIP have shown promising results, their capability in interpreting and following modification instructions remains limited. Some research attempts to address this by incorporating Large Language Models (LLMs). However, these approaches still face challenges in effectively integrating multimodal information and instruction understanding. To tackle above challenges, we propose a novel embedding method utilizing an instruction-tuned Multimodal LLM (MLLM) to generate composed representation, which significantly enhance the instruction following capability for a comprehensive integration between images and instructions. Nevertheless, directly applying MLLMs introduces a new challenge since MLLMs are primarily designed for text generation rather than embedding extraction as required in CIR. To address this, we introduce a two-stage training strategy to efficiently learn a joint multimodal embedding space and further refining the ability to follow modification instructions by tuning the model in a triplet dataset similar to the CIR format. Extensive experiments on four public datasets: FashionIQ, CIRR, GeneCIS, and CIRCO demonstrates the superior performance of our model, outperforming state-of-the-art baselines by a significant margin. Codes are available at the GitHub repository.
- Open Access
- 22 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-8
MFMF: Multiple Foundation Model Fusion Networks for Whole Slide Image Classification
Tumor detection and subtyping remain a significant challenge in histopathology image analysis. As digital pathology progresses, the applications of deep learning become essential. Whole Slide Image (WSI) classification has emerged as a crucial task in digital pathology, vital for accurate cancer diagnosis and treatment. In this paper, we introduce an innovative abnormal-guided Multiple Foundation Model Fusion (MFMF) framework, aimed at enhancing WSI classification by integrating multi-level information from pathology images with Multiple Instance Learning (MIL). Traditional methods often focus on patch-level features while neglecting the rich contextual and morphological details at the cell and text levels, thus failing to fully exploit the multidimensional nature of WSIs. Our method enhances traditional models by efficiently integrating patch-level, cell-level, and text-level features using three foundation models. These are then fused through a novel three-step cross-attention module that effectively leverages cell and text information with patch-level features. Furthermore, unlike most studies that use attention scores to select instances based on the assumption that high scores indicate the presence of a tumor, we design an abnormality-aware module to naturally identify and detect abnormal features (i.e., tumors) as the criteria for selecting important instances, thereby reducing computational costs and boosting overall performance. We validate our approach against leading benchmarks on the CAMELYON16 and TCGA-Lung datasets, achieving superior classification performance. Our study not only tackles the challenges of sparsity and noise in multi-level features but also enhances the efficiency and accuracy of WSI classification by exploiting abnormal features.
- 22 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-8
Open AccessMFMF: Multiple Foundation Model Fusion Networks for Whole Slide Image Classification
Tumor detection and subtyping remain a significant challenge in histopathology image analysis. As digital pathology progresses, the applications of deep learning become essential. Whole Slide Image (WSI) classification has emerged as a crucial task in digital pathology, vital for accurate cancer diagnosis and treatment. In this paper, we introduce an innovative abnormal-guided Multiple Foundation Model Fusion (MFMF) framework, aimed at enhancing WSI classification by integrating multi-level information from pathology images with Multiple Instance Learning (MIL). Traditional methods often focus on patch-level features while neglecting the rich contextual and morphological details at the cell and text levels, thus failing to fully exploit the multidimensional nature of WSIs. Our method enhances traditional models by efficiently integrating patch-level, cell-level, and text-level features using three foundation models. These are then fused through a novel three-step cross-attention module that effectively leverages cell and text information with patch-level features. Furthermore, unlike most studies that use attention scores to select instances based on the assumption that high scores indicate the presence of a tumor, we design an abnormality-aware module to naturally identify and detect abnormal features (i.e., tumors) as the criteria for selecting important instances, thereby reducing computational costs and boosting overall performance. We validate our approach against leading benchmarks on the CAMELYON16 and TCGA-Lung datasets, achieving superior classification performance. Our study not only tackles the challenges of sparsity and noise in multi-level features but also enhances the efficiency and accuracy of WSI classification by exploiting abnormal features.
- Open Access
- 22 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-8
AlphaEpi: Enhancing B Cell Epitope Prediction with AlphaFold 3
Accurately identifying conformational B-cell epitopes, which involve complex protein structures, is essential in modern immunology and vaccine development. Traditional methods like X-ray crystallography provide precise data but are costly and time-consuming. This has led to the rise of computational methods, especially deep learning algorithms, which offer more efficient and cost-effective solutions. A major challenge in predicting conformational epitopes based on structure is the limited availability of experimentally validated structural data. Structural prediction tools can help address this issue but still rely on provided sequence information and complex algorithms, which can introduce noise and affect accuracy. Moreover, effectively aligning and fusing structural features with evolutionary sequence information remains another challenge. To address these challenges, we introduce AlphaEpi, an innovative model that combines the advanced structure prediction capabilities of AlphaFold 3 with deep learning graph neural networks. Additionally, we propose for the first time a dynamic selector module that allows the model to adjust its processing strategy based on the reliability of the structure and sequence information, addressing the challenge posed by unreliable structural data. Moreover, a novel graph fusion module integrates structural and evolutionary information, effectively solving the disparities in the integration process of structure and sequence data. Results show that by integrating advanced structural prediction technologies AlphaFold 3 with deep learning graph neural networks, AlphaEpi significantly surpasses baseline results, establishing itself as the state-of-the-art. This advancement furthers the development of B-cell epitope prediction and has significant implications for biological research.
- 22 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-8
Open AccessAlphaEpi: Enhancing B Cell Epitope Prediction with AlphaFold 3
Accurately identifying conformational B-cell epitopes, which involve complex protein structures, is essential in modern immunology and vaccine development. Traditional methods like X-ray crystallography provide precise data but are costly and time-consuming. This has led to the rise of computational methods, especially deep learning algorithms, which offer more efficient and cost-effective solutions. A major challenge in predicting conformational epitopes based on structure is the limited availability of experimentally validated structural data. Structural prediction tools can help address this issue but still rely on provided sequence information and complex algorithms, which can introduce noise and affect accuracy. Moreover, effectively aligning and fusing structural features with evolutionary sequence information remains another challenge. To address these challenges, we introduce AlphaEpi, an innovative model that combines the advanced structure prediction capabilities of AlphaFold 3 with deep learning graph neural networks. Additionally, we propose for the first time a dynamic selector module that allows the model to adjust its processing strategy based on the reliability of the structure and sequence information, addressing the challenge posed by unreliable structural data. Moreover, a novel graph fusion module integrates structural and evolutionary information, effectively solving the disparities in the integration process of structure and sequence data. Results show that by integrating advanced structural prediction technologies AlphaFold 3 with deep learning graph neural networks, AlphaEpi significantly surpasses baseline results, establishing itself as the state-of-the-art. This advancement furthers the development of B-cell epitope prediction and has significant implications for biological research.
- Open Access
- 12 June 2024
- Published by MDPI AG in BioMedInformatics
- Vol. 4 (2) , 1556-1571
Advancing DNA Language Models through Motif-Oriented Pre-Training with MoDNA
Acquiring meaningful representations of gene expression is essential for the accurate prediction of downstream regulatory tasks, such as identifying promoters and transcription factor binding sites. However, the current dependency on supervised learning, constrained by the limited availability of labeled genomic data, impedes the ability to develop robust predictive models with broad generalization capabilities. In response, recent advancements have pivoted towards the application of self-supervised training for DNA sequence modeling, enabling the adaptation of pre-trained genomic representations to a variety of downstream tasks. Departing from the straightforward application of masked language learning techniques to DNA sequences, approaches such as MoDNA enrich genome language modeling with prior biological knowledge. In this study, we advance DNA language models by utilizing the Motif-oriented DNA (MoDNA) pre-training framework, which is established for self-supervised learning at the pre-training stage and is flexible enough for application across different downstream tasks. MoDNA distinguishes itself by efficiently learning semantic-level genomic representations from an extensive corpus of unlabeled genome data, offering a significant improvement in computational efficiency over previous approaches. The framework is pre-trained on a comprehensive human genome dataset and fine-tuned for targeted downstream tasks. Our enhanced analysis and evaluation in promoter prediction and transcription factor binding site prediction have further validated MoDNA’s exceptional capabilities, emphasizing its contribution to advancements in genomic predictive modeling.
- 12 June 2024
- Published by MDPI AG in BioMedInformatics
- Vol. 4 (2) , 1556-1571
Open AccessAdvancing DNA Language Models through Motif-Oriented Pre-Training with MoDNA
Acquiring meaningful representations of gene expression is essential for the accurate prediction of downstream regulatory tasks, such as identifying promoters and transcription factor binding sites. However, the current dependency on supervised learning, constrained by the limited availability of labeled genomic data, impedes the ability to develop robust predictive models with broad generalization capabilities. In response, recent advancements have pivoted towards the application of self-supervised training for DNA sequence modeling, enabling the adaptation of pre-trained genomic representations to a variety of downstream tasks. Departing from the straightforward application of masked language learning techniques to DNA sequences, approaches such as MoDNA enrich genome language modeling with prior biological knowledge. In this study, we advance DNA language models by utilizing the Motif-oriented DNA (MoDNA) pre-training framework, which is established for self-supervised learning at the pre-training stage and is flexible enough for application across different downstream tasks. MoDNA distinguishes itself by efficiently learning semantic-level genomic representations from an extensive corpus of unlabeled genome data, offering a significant improvement in computational efficiency over previous approaches. The framework is pre-trained on a comprehensive human genome dataset and fine-tuned for targeted downstream tasks. Our enhanced analysis and evaluation in promoter prediction and transcription factor binding site prediction have further validated MoDNA’s exceptional capabilities, emphasizing its contribution to advancements in genomic predictive modeling.
- 1 March 2024
- Published by SAGE Publications in Journal of Computational Biology
- Vol. 31 (3) , 213-228
Toward Robust Self-Training Paradigm for Molecular Prediction Tasks
Molecular prediction tasks normally demand a series of professional experiments to label the target molecule, which suffers from the limited labeled data problem. One of the semisupervised learning paradigms, known as self-training, utilizes both labeled and unlabeled data. Specifically, a teacher model is trained using labeled data and produces pseudo labels for unlabeled data. These labeled and pseudo-labeled data are then jointly used to train a student model. However, the pseudo labels generated from the teacher model are generally not sufficiently accurate. Thus, we propose a robust self-training strategy by exploring robust loss function to handle such noisy labels in two paradigms, that is, generic and adaptive. We have conducted experiments on three molecular biology prediction tasks with four backbone models to gradually evaluate the performance of the proposed robust self-training strategy. The results demonstrate that the proposed method enhances prediction performance across all tasks, notably within molecular regression tasks, where there has been an average enhancement of 41.5%. Furthermore, the visualization analysis confirms the superiority of our method. Our proposed robust self-training is a simple yet effective strategy that efficiently improves molecular biology prediction performance. It tackles the labeled data insufficient issue in molecular biology by taking advantage of both labeled and unlabeled data. Moreover, it can be easily embedded with any prediction task, which serves as a universal approach for the bioinformatics community.
- 1 March 2024
- Published by SAGE Publications in Journal of Computational Biology
- Vol. 31 (3) , 213-228
Toward Robust Self-Training Paradigm for Molecular Prediction Tasks
Molecular prediction tasks normally demand a series of professional experiments to label the target molecule, which suffers from the limited labeled data problem. One of the semisupervised learning paradigms, known as self-training, utilizes both labeled and unlabeled data. Specifically, a teacher model is trained using labeled data and produces pseudo labels for unlabeled data. These labeled and pseudo-labeled data are then jointly used to train a student model. However, the pseudo labels generated from the teacher model are generally not sufficiently accurate. Thus, we propose a robust self-training strategy by exploring robust loss function to handle such noisy labels in two paradigms, that is, generic and adaptive. We have conducted experiments on three molecular biology prediction tasks with four backbone models to gradually evaluate the performance of the proposed robust self-training strategy. The results demonstrate that the proposed method enhances prediction performance across all tasks, notably within molecular regression tasks, where there has been an average enhancement of 41.5%. Furthermore, the visualization analysis confirms the superiority of our method. Our proposed robust self-training is a simple yet effective strategy that efficiently improves molecular biology prediction performance. It tackles the labeled data insufficient issue in molecular biology by taking advantage of both labeled and unlabeled data. Moreover, it can be easily embedded with any prediction task, which serves as a universal approach for the bioinformatics community.
- Open Access
- 2 June 2022
- Published by MDPI AG in Biomolecules
- Vol. 12 (6) , 774
Deep Ensemble Learning with Atrous Spatial Pyramid Networks for Protein Secondary Structure Prediction
The secondary structure of proteins is significant for studying the three-dimensional structure and functions of proteins. Several models from image understanding and natural language modeling have been successfully adapted in the protein sequence study area, such as Long Short-term Memory (LSTM) network and Convolutional Neural Network (CNN). Recently, Gated Convolutional Neural Network (GCNN) has been proposed for natural language processing. It has achieved high levels of sentence scoring, as well as reduced the latency. Conditionally Parameterized Convolution (CondConv) is another novel study which has gained great success in the image processing area. Compared with vanilla CNN, CondConv uses extra sample-dependant modules to conditionally adjust the convolutional network. In this paper, we propose a novel Conditionally Parameterized Convolutional network (CondGCNN) which utilizes the power of both CondConv and GCNN. CondGCNN leverages an ensemble encoder to combine the capabilities of both LSTM and CondGCNN to encode protein sequences by better capturing protein sequential features. In addition, we explore the similarity between the secondary structure prediction problem and the image segmentation problem, and propose an ASP network (Atrous Spatial Pyramid Pooling (ASPP) based network) to capture fine boundary details in secondary structure. Extensive experiments show that the proposed method can achieve higher performance on protein secondary structure prediction task than existing methods on CB513, Casp11, CASP12, CASP13, and CASP14 datasets. We also conducted ablation studies over each component to verify the effectiveness. Our method is expected to be useful for any protein related prediction tasks, which is not limited to protein secondary structure prediction.
- 2 June 2022
- Published by MDPI AG in Biomolecules
- Vol. 12 (6) , 774
Open AccessDeep Ensemble Learning with Atrous Spatial Pyramid Networks for Protein Secondary Structure Prediction
The secondary structure of proteins is significant for studying the three-dimensional structure and functions of proteins. Several models from image understanding and natural language modeling have been successfully adapted in the protein sequence study area, such as Long Short-term Memory (LSTM) network and Convolutional Neural Network (CNN). Recently, Gated Convolutional Neural Network (GCNN) has been proposed for natural language processing. It has achieved high levels of sentence scoring, as well as reduced the latency. Conditionally Parameterized Convolution (CondConv) is another novel study which has gained great success in the image processing area. Compared with vanilla CNN, CondConv uses extra sample-dependant modules to conditionally adjust the convolutional network. In this paper, we propose a novel Conditionally Parameterized Convolutional network (CondGCNN) which utilizes the power of both CondConv and GCNN. CondGCNN leverages an ensemble encoder to combine the capabilities of both LSTM and CondGCNN to encode protein sequences by better capturing protein sequential features. In addition, we explore the similarity between the secondary structure prediction problem and the image segmentation problem, and propose an ASP network (Atrous Spatial Pyramid Pooling (ASPP) based network) to capture fine boundary details in secondary structure. Extensive experiments show that the proposed method can achieve higher performance on protein secondary structure prediction task than existing methods on CB513, Casp11, CASP12, CASP13, and CASP14 datasets. We also conducted ablation studies over each component to verify the effectiveness. Our method is expected to be useful for any protein related prediction tasks, which is not limited to protein secondary structure prediction.
- 1 April 2021
- Published by SAGE Publications in Journal of Computational Biology
- Vol. 28 (4) , 346-361
Comprehensive Study on Enhancing Low-Quality Position-Specific Scoring Matrix with Deep Learning for Accurate Protein Structure Property Prediction: Using Bagging Multiple Sequence Alignment Learning
Accurate predictions of protein structure properties, for example, secondary structure and solvent accessibility, are essential in analyzing the structure and function of a protein. Position-specific scoring matrix (PSSM) features are widely used in the structure property prediction. However, some proteins may have low-quality PSSM features due to insufficient homologous sequences, leading to limited prediction accuracy. To address this limitation, we propose an enhancing scheme for PSSM features. We introduce the “Bagging MSA” (multiple sequence alignment) method to calculate PSSM features used to train our model, adopt a convolutional network to capture local context features and bidirectional long short-term memory for long-term dependencies, and integrate them under an unsupervised framework. Structure property prediction models are then built upon such enhanced PSSM features for more accurate predictions. Moreover, we develop two frameworks to evaluate the effectiveness of the enhanced PSSM features, which also bring proposed method into real-world scenarios. Empirical evaluation of CB513, CASP11, and CASP12 data sets indicates that our unsupervised enhancing scheme indeed generates more informative PSSM features for structure property prediction.
- 1 April 2021
- Published by SAGE Publications in Journal of Computational Biology
- Vol. 28 (4) , 346-361
Comprehensive Study on Enhancing Low-Quality Position-Specific Scoring Matrix with Deep Learning for Accurate Protein Structure Property Prediction: Using Bagging Multiple Sequence Alignment Learning
Accurate predictions of protein structure properties, for example, secondary structure and solvent accessibility, are essential in analyzing the structure and function of a protein. Position-specific scoring matrix (PSSM) features are widely used in the structure property prediction. However, some proteins may have low-quality PSSM features due to insufficient homologous sequences, leading to limited prediction accuracy. To address this limitation, we propose an enhancing scheme for PSSM features. We introduce the “Bagging MSA” (multiple sequence alignment) method to calculate PSSM features used to train our model, adopt a convolutional network to capture local context features and bidirectional long short-term memory for long-term dependencies, and integrate them under an unsupervised framework. Structure property prediction models are then built upon such enhanced PSSM features for more accurate predictions. Moreover, we develop two frameworks to evaluate the effectiveness of the enhanced PSSM features, which also bring proposed method into real-world scenarios. Empirical evaluation of CB513, CASP11, and CASP12 data sets indicates that our unsupervised enhancing scheme indeed generates more informative PSSM features for structure property prediction.