Youngmahn Han
Research Subjects
Youngmahn Han has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- Kiseong Kim
The R&D Center, BioBrain Inc., Daejeon 35203, Republic of Korea
2 shared publications Aeri Lee
BioBrain Inc
1 shared publication- Jin Yong Lee
Department of Chemistry, Sungkyunkwan University, Suwon 16419, Korea
1 shared publication - Jong Hyeon Lim
Department of Chemistry Sungkyunkwan University Suwon Republic of Korea
1 shared publication - Kavya Dasaramoole Prakash
Bio-AI R&D center, AIdanBio Inc, Daejeon, Republic of Korea
1 shared publication - Myeonghyeon Jeong
Department of ICT Convergence System Engineering, Chonnam National University, Gwangju, South Korea
1 shared publication
Latest Publications
- Open Access
Enhancing Clinical Reasoning in Medical Vision-Language Model through Structured Prompts
Medical Vision-Language Models (MVLMs) are emerging as powerful tools for tasks such as Visual Question Answering (VQA); however, they often struggle with hallucination and limited reasoning transparency, particularly in complex diagnostic scenarios. In this work, we enhance the MedVLM-R1 framework by fine-tuning it using clinically informed prompt structures tailored specifically for radiology-based reasoning. Without altering the original model architecture or training strategy, we redesign the system prompts and question templates to guide the model through structured, modality-aware, and step-by-step diagnostic reasoning. Fine-tuning is performed using MRI-based question-answer (QA) pairs, and evaluations are conducted across three diagnostic imaging: MRI, CT, and X-ray to assess both in-domain and out-of-domain generalization. Our approach improves reasoning transparency and accuracy, achieving 96.00% on MRI, 72.67% on CT, and 75.2% on X-ray. Compared to the original MedVLM-R1, our method closes the gap in MRI accuracy while significantly enhancing generalization performance on CT and X-ray modalities. These results demonstrate that clinically grounded prompting effectively improves both reasoning fidelity and robustness across imaging modalities. The code is available at our GitHub repository: https://github.com/aidanbio/AIdanMed
Open AccessEnhancing Clinical Reasoning in Medical Vision-Language Model through Structured Prompts
Medical Vision-Language Models (MVLMs) are emerging as powerful tools for tasks such as Visual Question Answering (VQA); however, they often struggle with hallucination and limited reasoning transparency, particularly in complex diagnostic scenarios. In this work, we enhance the MedVLM-R1 framework by fine-tuning it using clinically informed prompt structures tailored specifically for radiology-based reasoning. Without altering the original model architecture or training strategy, we redesign the system prompts and question templates to guide the model through structured, modality-aware, and step-by-step diagnostic reasoning. Fine-tuning is performed using MRI-based question-answer (QA) pairs, and evaluations are conducted across three diagnostic imaging: MRI, CT, and X-ray to assess both in-domain and out-of-domain generalization. Our approach improves reasoning transparency and accuracy, achieving 96.00% on MRI, 72.67% on CT, and 75.2% on X-ray. Compared to the original MedVLM-R1, our method closes the gap in MRI accuracy while significantly enhancing generalization performance on CT and X-ray modalities. These results demonstrate that clinically grounded prompting effectively improves both reasoning fidelity and robustness across imaging modalities. The code is available at our GitHub repository: https://github.com/aidanbio/AIdanMed
- Open Access
- 21 July 2025
- Published by American Chemical Society (ACS) in ChemRxiv
DeBERTa-Based SMILES Encoders for ADMET-Aware Drug Design
Multi-modal drug discovery frameworks increasingly rely on robust encoders capable of representing chemical structures alongside other data modalities. In this study, we fine-tuned a DeBERTa-based SMILES encoder to improve its predictive capacity for 22 ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) endpoints while maintaining strong comprehension of molecular syntax. Starting from a pretrained ZINC-based DeBERTa checkpoint, we trained on a 300K PubChem-ADMET dataset using a multi-label regression scheme with a focal MAE loss to stabilize learning across diverse properties. Our encoder achieved top-10 rankings on 16 TDC benchmark tasks, including notable improvements of 14–30% on critical endpoints such as bioavailability and CYP2C9-substrate. Compared with BERT- and RoBERTa-based molecular encoders, our approach preserved significantly higher MLM accuracy (>89%) over training, indicating robust retention of chemical language understanding. Additional analysis using an ADMET path length metric revealed that DeBERTa produced more disentangled latent representations, underscoring its suitability for property-specific molecular manipulation. These results demonstrate that a disentangled, ADMET-aware DeBERTa encoder can serve as a powerful component for future multi-modal pipelines in AI-driven drug design, effectively balancing structural fluency with predictive specialization.
- 21 July 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessDeBERTa-Based SMILES Encoders for ADMET-Aware Drug Design
Multi-modal drug discovery frameworks increasingly rely on robust encoders capable of representing chemical structures alongside other data modalities. In this study, we fine-tuned a DeBERTa-based SMILES encoder to improve its predictive capacity for 22 ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) endpoints while maintaining strong comprehension of molecular syntax. Starting from a pretrained ZINC-based DeBERTa checkpoint, we trained on a 300K PubChem-ADMET dataset using a multi-label regression scheme with a focal MAE loss to stabilize learning across diverse properties. Our encoder achieved top-10 rankings on 16 TDC benchmark tasks, including notable improvements of 14–30% on critical endpoints such as bioavailability and CYP2C9-substrate. Compared with BERT- and RoBERTa-based molecular encoders, our approach preserved significantly higher MLM accuracy (>89%) over training, indicating robust retention of chemical language understanding. Additional analysis using an ADMET path length metric revealed that DeBERTa produced more disentangled latent representations, underscoring its suitability for property-specific molecular manipulation. These results demonstrate that a disentangled, ADMET-aware DeBERTa encoder can serve as a powerful component for future multi-modal pipelines in AI-driven drug design, effectively balancing structural fluency with predictive specialization.
- Open Access
- 21 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/ACM Transactions on Computational Biology and Bioinformatics
- Vol. 21 (3) , 428-438
Interpretable Prediction of SARS-CoV-2 Epitope-Specific TCR Recognition Using a Pre-Trained Protein Language Model
The emergence of the novel coronavirus, designated as severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2), has posed a significant threat to public health worldwide. There has been progress in reducing hospitalizations and deaths due to SARS-CoV-2. However, challenges stem from the emergence of SARS-CoV-2 variants, which exhibit high transmission rates, increased disease severity, and the ability to evade humoral immunity. Epitope-specific T-cell receptor (TCR) recognition is key in determining the T-cell immunogenicity for SARS-CoV-2 epitopes. Although several data-driven methods for predicting epitope-specific TCR recognition have been proposed, they remain challenging due to the enormous diversity of TCRs and the lack of available training data. Self-supervised transfer learning has recently been proven useful for extracting information from unlabeled protein sequences, increasing the predictive performance of fine-tuned models, and using a relatively small amount of training data. This study presents a deep-learning model generated by fine-tuning pre-trained protein embeddings from a large corpus of protein sequences. The fine-tuned model showed markedly high predictive performance and outperformed the recent Gaussian process-based prediction model. The output attentions captured by the deep-learning model suggested critical amino acid positions in the SARS-CoV-2 epitope-specific TCRβ sequences that are highly associated with the viral escape of T-cell immune response.
- 21 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/ACM Transactions on Computational Biology and Bioinformatics
- Vol. 21 (3) , 428-438
Open AccessInterpretable Prediction of SARS-CoV-2 Epitope-Specific TCR Recognition Using a Pre-Trained Protein Language Model
The emergence of the novel coronavirus, designated as severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2), has posed a significant threat to public health worldwide. There has been progress in reducing hospitalizations and deaths due to SARS-CoV-2. However, challenges stem from the emergence of SARS-CoV-2 variants, which exhibit high transmission rates, increased disease severity, and the ability to evade humoral immunity. Epitope-specific T-cell receptor (TCR) recognition is key in determining the T-cell immunogenicity for SARS-CoV-2 epitopes. Although several data-driven methods for predicting epitope-specific TCR recognition have been proposed, they remain challenging due to the enormous diversity of TCRs and the lack of available training data. Self-supervised transfer learning has recently been proven useful for extracting information from unlabeled protein sequences, increasing the predictive performance of fine-tuned models, and using a relatively small amount of training data. This study presents a deep-learning model generated by fine-tuning pre-trained protein embeddings from a large corpus of protein sequences. The fine-tuned model showed markedly high predictive performance and outperformed the recent Gaussian process-based prediction model. The output attentions captured by the deep-learning model suggested critical amino acid positions in the SARS-CoV-2 epitope-specific TCRβ sequences that are highly associated with the viral escape of T-cell immune response.
- Open Access
Predicting SARS-CoV-2 epitope-specific TCR recognition using pre-trained protein embeddings
The COVID-19 pandemic is ongoing because of the high transmission rate and the emergence of SARS-CoV-2 variants. The P272L mutation in SARS-Cov-2 S -protein is known to be highly relevant to the viral escape associated with the second pandemic wave in Europe. Epitope-specific T-cell receptor (TCR) recognition is a key factor in determining the T-cell immunogenicity of a SARS-CoV-2 epitope. Although several data-driven methods for predicting epitope-specific TCR recognition have been proposed, they remain challenging owing to the enormous diversity of TCRs and the lack of available training data. Self-supervised transfer learning has recently been demonstrated to be powerful for extracting useful information from unlabeled protein sequences and increasing the predictive performance of the fine-tuned models in downstream tasks. Here, we present a predictive model based on Bidirectional Encoder Representations from Transformers (BERT), employing self-supervised transfer learning, to predict SARS-CoV-2 T-cell epitope-specific TCR recognition. The fine-tuned model showed notably high predictive performance for independent evaluation using the SARS-CoV-2 epitope-specific TCR CDR3β sequence datasets. In particular, we found the proline at position 4 corresponding to the P272L mutation in the SARS-CoV-2 S -protein 269-277 epitope ( YLQPRTFLL ) may contribute substantially to TCR recognition of the epitope through interpreting the output attention weights of our model. We anticipate that our findings will provide new directions for constructing a reliable data-driven model to predict the immunogenic T-cell epitopes using limited training data and help accelerate the development of an effective vaccine in response to SARS-CoV-2 variants.
Open AccessPredicting SARS-CoV-2 epitope-specific TCR recognition using pre-trained protein embeddings
The COVID-19 pandemic is ongoing because of the high transmission rate and the emergence of SARS-CoV-2 variants. The P272L mutation in SARS-Cov-2 S -protein is known to be highly relevant to the viral escape associated with the second pandemic wave in Europe. Epitope-specific T-cell receptor (TCR) recognition is a key factor in determining the T-cell immunogenicity of a SARS-CoV-2 epitope. Although several data-driven methods for predicting epitope-specific TCR recognition have been proposed, they remain challenging owing to the enormous diversity of TCRs and the lack of available training data. Self-supervised transfer learning has recently been demonstrated to be powerful for extracting useful information from unlabeled protein sequences and increasing the predictive performance of the fine-tuned models in downstream tasks. Here, we present a predictive model based on Bidirectional Encoder Representations from Transformers (BERT), employing self-supervised transfer learning, to predict SARS-CoV-2 T-cell epitope-specific TCR recognition. The fine-tuned model showed notably high predictive performance for independent evaluation using the SARS-CoV-2 epitope-specific TCR CDR3β sequence datasets. In particular, we found the proline at position 4 corresponding to the P272L mutation in the SARS-CoV-2 S -protein 269-277 epitope ( YLQPRTFLL ) may contribute substantially to TCR recognition of the epitope through interpreting the output attention weights of our model. We anticipate that our findings will provide new directions for constructing a reliable data-driven model to predict the immunogenic T-cell epitopes using limited training data and help accelerate the development of an effective vaccine in response to SARS-CoV-2 variants.