Saiyang Na

Experienced Computer Science Ph.D. candidate with expertise in deep learning and LLM. Created multiple open-source libraries with JAX and Python, focusing on deep learning and LLM.

Education

Publications

• Saiyang Na, Junzhou Huang, (2026), “Deep learning for TCR–pMHC binding prediction”, Deep Learning in Drug Design, 381-402.
• Saiyang Na, Bing He, Jie Liu, Siyang Jia, Fatemeh Tahsin, Fang Zhang, Hao Chen, Junzhou Huang, (2025), “Segment Any Cell: A SAM-based Auto-prompting Fine-tuning Framework for Nuclei Segmentation”, IEEE TNNLS.
• Bing Song, Kaiwen Wang, Saiyang Na, Haochen Shi, Xiaobing Han, Linqi Dai, Zeyue Yang, Tianwei Yu, Junzhou Huang, (2025), “Profiling antigen-binding affinity of B cell repertoires in tumors by deep learning predicts immune-checkpoint inhibitor treatment outcomes”, Nature Cancer 6, 1570-1584.
• Thao M Dang, Qifeng Zhou, Yuzhi Guo, Hehuan Ma, Saiyang Na, TB Dang, Jian Gao, Junzhou Huang, (2025), “Abnormality-aware multimodal learning for WSI classification”, Front. Med. 12, 1546452.
• Qifeng Zhou, Thao M Dang, Wenliang Zhong, Yuzhi Guo, Hehuan Ma, Saiyang Na, Hao Li, Junzhou Huang, (2025), “MLLM4PUE: Toward Universal Embeddings in Digital Pathology through Multimodal LLMs”, arXiv:2502.07221.
• Qifeng Zhou, Xinyu Zhu, Saiyang Na, Christi Alappat, Jacob Antony Secakusuma, Hawk Weisman, Jingyi Hu, Vu Nguyen, Pushkar Khetrapal, Yuzhi Guo, Junzhou Huang, (2025), “Contrastive Pretraining for Computational Pathology with Visual-Language Models”, IEEE ISBI.
• Feng Jiang, Saiyang Na, Yuzhi Guo, Wenliang Zhong, Yikang Han, Tao Wang, Junzhou Huang, (2024), “AlphaEpi: Enhancing B Cell Epitope Prediction with AlphaFold 3”, ACM BCB.
• Thao M Dang, Qifeng Zhou, Christi Alappat, Saiyang Na, Yuzhi Guo, Pushkar Khetrapal, Vu Nguyen, Xiuzhen Huang, Junzhou Huang, (2024), “MFMF: multiple foundation model fusion networks for whole slide image classification”, ACM BCB.
• Feng Jiang, Yuzhi Guo, Hehuan Ma, Saiyang Na, Wenliang Zhong, Yikang Han, Tao Wang, Junzhou Huang, (2024), “GTE: a graph learning framework for prediction of T-cell receptors and epitopes binding specificity”, Brief. Bioinform. 25 (4), bbae343.
• Lu Zhang, Saiyang Na, Tianming Liu, Dajiang Zhu, Junzhou Huang, (2023), “Multimodal deep fusion in hyperbolic space for mild cognitive impairment study”, MICCAI, Oral.
• Xinyue Ye, Jiaxin Du, Xi Gong, Saiyang Na, Yifei Sun, Jay Lee, (2021), “Geospatial and semantic mapping platform for massive COVID-19 scientific publication search”, J. Geoviz. Spat. Anal. 5 (1), 1-15.

Experience

• Developed HyperGRAM, extending Gramian volume-based multimodal alignment to hyperbolic space using Lorentz model.
• Achieved SOTA zero-shot performance on video-text retrieval: MSR-VTT (56.6%), ActivityNet (58.2%), VATEX (79.9%).

• Achieved state-of-the-art performance in multi-step retrosynthesis on the USPTO dataset by combining GNN and LLM, building upon the LocalRetro framework.
• Participated in the Standard Industries Chemical Innovation Challenge on HeroX, a retrosynthesis competition, and advanced to the semi-finals, placing in the top 10.
• Extended research to the Pistachio dataset for industrial-scale retrosynthesis prediction.

• Developed auto-prompt generation pipeline for SAM achieving SOTA performance in cell segmentation.
• Created live demo at https://segment-any-cell.com.
• Published in IEEE TNNLS (2025).

• Developed contrastive learning framework for TCR-pMHC binding prediction with SOTA accuracy.
• Built end-to-end protein sequence-to-structure pipeline integrating RoseTTAFold.
• Published in top journals including Nature Cancer (2025) and Brief. Bioinform. (2024).
• Authored book chapter on deep learning for TCR-pMHC binding prediction in Deep Learning in Drug Design (2026).
• Open-source implementations: GTE (https://github.com/uta-smile/GTE) and Cmai (https://github.com/ice4prince/Cmai).

• Applied hyperbolic graph neural networks to MCI classification, achieving 92.30% accuracy.
• Published in MICCAI 2023 (Oral presentation). Source code: https://github.com/nasyxx/MDF-HS.
• Developed JAX-based hyperbolic neural networks library: https://github.com/nasyxx/jaxrie.

• Performed zebrafish juvenile heart segmentation using nnNet.
• Code: https://github.com/nasyxx/zebrafish_seg

• Analysis of the evolution process of Heterogeneous Graphs in deep learning, on the graph-related tasks, such as node classification tasks.

• Implemented 2D/3D UNet architectures with PyTorch and TensorFlow.
• Developed efficient data pipeline reducing processing time.
• Created TPU-optimized data augmentation techniques for medical images.

• Source: https://github.com/nasyxx/cord_search
• Related paper:
  • Xinyue Ye et al., “Geospatial and semantic mapping platform for massive COVID-19 scientific publication search”, Journal of Geovisualization and Spatial Analysis
  • https://doi.org/10.1007/s41651-021-00073-y
• Build a web app to search for covid-19 related papers.

Open-Source Projects

• A deep learning framework based on JAX and Equinox, which includes custom model implementations, JAX and GPU-compatible dataloaders, an Equinox trainer, and various JAX-based helper utilities.

• A JAX hyperbolic neural networks library.

• An IPython Extensions, which includes a better trace exception and auto performance process and CPU timer.

Teaching and Researching Assistance

• Research Assistance of Dr. Xinyue Ye, 2019 to 2020
• Research Assistance of Dr. Junzhou Huang, start from 2025
• CSE1310, Introduction to Computers & Programming, Fall 2021, Spring 2022, Fall 2022
• CSE5311, Design and Analysis of Algorithms, Fall 2023, Fall 2024
• CSE5324, Software Engineering, Summer 2022
• CSE5334, Data Mining, Summer 2024
• CSE6392, Special Topics in Deep Learning, Spring 2024

Skills

• Expert: Python, JAX with Equinox, PyTorch, NumPy
• Proficient: TensorFlow with Keras, Lisp (Emacs Lisp), LLM
• Familiar: Haskell, JavaScript, C/C++

Internship

• Our team mapped the route and found the coincident points. We analyzed the right path, corrected the real-time direction to match the track and the bus's designated route, and realized the bus's real-time position.

• Participated in Establishing real time advertisement/commercial bidding (RTB) model.
• Built target people labeling system and made classification with logistic regression.
• The model was well applied into company's practices.