Objectives: Developing large language models (LLMs) in biomedicine requires access to high-quality training and alignment tuning datasets. However, publicly available Korean medical preference datasets are scarce, hindering the advancement of Korean medical LLMs. This study constructs and evaluates the efficacy of the Korean Medical Preference Dataset (KoMeP), an alignment tuning dataset constructed with an automated pipeline, minimizing the high costs of human annotation. Methods: KoMeP was generated using the DAHL score, an automated hallucination evaluation metric. Five LLMs (Dolly-v2-3B, MPT-7B, GPT-4o, Qwen-2-7B, Llama-3-8B) produced responses to 8,573 biomedical examination questions, from which 5,551 preference pairs were extracted. Each pair consisted of a “chosen” response and a “rejected” response, as determined by their DAHL scores. The dataset was evaluated when trained through two different alignment tuning methods, direct preference optimization (DPO) and odds ratio preference optimization (ORPO) respectively across five different models. The KorMedMCQA benchmark was employed to assess the effectiveness of alignment tuning. Results: Models trained with DPO consistently improved KorMedMCQA performance; notably, Llama-3.1-8B showed a 43.96% increase. In contrast, ORPO training produced inconsistent results. Additionally, English-to-Korean transfer learning proved effective, particularly for English-centric models like Gemma-2, whereas Korean-to-English transfer learning achieved limited success. Instruction tuning with KoMeP yielded mixed outcomes, which suggests challenges in dataset formatting. Conclusions KoMeP is the first publicly available Korean medical preference dataset and significantly improves alignment tuning performance in LLMs. The DPO method outperforms ORPO in alignment tuning. Future work should focus on expanding KoMeP, developing a Korean-native dataset, and refining alignment tuning methods to produce safer and more reliable Korean medical LLMs.

Thelazia callipaeda (Spirurida, Thelaziidae), a parasitic eye worm causing thelaziasis has been reported in humans and dogs in Korea. However, its occurrence in other potential reservoir hosts, including wild animals, remains unclear. In the present study, we described the 2 cases of thelaziasis from both of feral and captive wild animal in Korea. In August and November 2017, 2 cases of the parasitic infections were found in the third eyelid of rescued leopard cat Prionailurus bengalensis and reared raccoon dog Nyctereutes procyonoides at the Chungnam Wild Animal Rescue Center. A total of 20 and 24 worms were detected from the left and right eyes of leopard cat, respectively. In the left eye of the raccoon dog, 5 worms were recognized. Male worms were 969–11,860 μm long (10,600 μm on average) and 300–320 μm width (315 μm on average). Female worms were 13,430–15,330 (14,480) μm long and 320–370 (344) μm wide in size. They commonly had a characteristic scalariform buccal cavity and short esophagus. The vulva openings were located at the anterior of esophago-intestinal junction in females. The thelaziasis is reported in 2 species of wildlife, P. bengalensis and N. procyonoides, for the first time in Korea.

Thelazia callipaeda (Spirurida, Thelaziidae), a parasitic eye worm causing thelaziasis has been reported in humans and dogs in Korea. However, its occurrence in other potential reservoir hosts, including wild animals, remains unclear. In the present study, we described the 2 cases of thelaziasis from both of feral and captive wild animal in Korea. In August and November 2017, 2 cases of the parasitic infections were found in the third eyelid of rescued leopard cat Prionailurus bengalensis and reared raccoon dog Nyctereutes procyonoides at the Chungnam Wild Animal Rescue Center. A total of 20 and 24 worms were detected from the left and right eyes of leopard cat, respectively. In the left eye of the raccoon dog, 5 worms were recognized. Male worms were 969–11,860 μm long (10,600 μm on average) and 300–320 μm width (315 μm on average). Female worms were 13,430–15,330 (14,480) μm long and 320–370 (344) μm wide in size. They commonly had a characteristic scalariform buccal cavity and short esophagus. The vulva openings were located at the anterior of esophago-intestinal junction in females. The thelaziasis is reported in 2 species of wildlife, P. bengalensis and N. procyonoides, for the first time in Korea.

Background: Circulating tumor DNA (ctDNA) is a potential biomarker in pancreatic ductal adenocarcinoma (PDAC). However, studies on residual ctDNA in patients post-chemotherapy are limited. We assessed the prognostic value of residual ctDNA in metastatic PDAC relative to that of carbohydrate antigen 19-9 (CA19-9). Methods: ctDNA analysis using a targeted next-generation sequencing panel was performed at baseline and during chemotherapy response evaluation in 53 patients. Progression-free survival (PFS) and overall survival (OS) were first evaluated based on ctDNA positivity at baseline. For further comparison, patients testing ctDNA-positive at baseline were subdivided based on residual ctDNA into ctDNA responders (no residual ctDNA post-chemotherapy) and ctDNA non-responders (residual ctDNA post-chemotherapy). Additional survival analysis was performed based on CA19-9 levels. Results: The baseline ctDNA detection rate was 56.6%. Although clinical outcomes tended to be poorer in patients with baseline ctDNA positivity than in those without, the differences were not significant. Residual ctDNA post-chemotherapy was associated with reduced PFS and OS. The prognosis of ctDNA responders was better than that of non-responders but did not significantly differ from that of ctDNA-negative individuals (no ctDNA both at baseline and during post-chemotherapy). Compared with ctDNA responses to che-motherapy, a ≥ 50% decrease in the CA19-9 level had less effect on both PFS and OSbased on hazard ratios and significance levels. ctDNA could be monitored in half of the patients whose baseline CA19-9 levels were within the reference range. Conclusions Residual ctDNA analysis post-chemotherapy is a promising approach for predicting the clinical outcomes of patients with metastatic PDAC.

Objectives: Developing large language models (LLMs) in biomedicine requires access to high-quality training and alignment tuning datasets. However, publicly available Korean medical preference datasets are scarce, hindering the advancement of Korean medical LLMs. This study constructs and evaluates the efficacy of the Korean Medical Preference Dataset (KoMeP), an alignment tuning dataset constructed with an automated pipeline, minimizing the high costs of human annotation. Methods: KoMeP was generated using the DAHL score, an automated hallucination evaluation metric. Five LLMs (Dolly-v2-3B, MPT-7B, GPT-4o, Qwen-2-7B, Llama-3-8B) produced responses to 8,573 biomedical examination questions, from which 5,551 preference pairs were extracted. Each pair consisted of a “chosen” response and a “rejected” response, as determined by their DAHL scores. The dataset was evaluated when trained through two different alignment tuning methods, direct preference optimization (DPO) and odds ratio preference optimization (ORPO) respectively across five different models. The KorMedMCQA benchmark was employed to assess the effectiveness of alignment tuning. Results: Models trained with DPO consistently improved KorMedMCQA performance; notably, Llama-3.1-8B showed a 43.96% increase. In contrast, ORPO training produced inconsistent results. Additionally, English-to-Korean transfer learning proved effective, particularly for English-centric models like Gemma-2, whereas Korean-to-English transfer learning achieved limited success. Instruction tuning with KoMeP yielded mixed outcomes, which suggests challenges in dataset formatting. Conclusions KoMeP is the first publicly available Korean medical preference dataset and significantly improves alignment tuning performance in LLMs. The DPO method outperforms ORPO in alignment tuning. Future work should focus on expanding KoMeP, developing a Korean-native dataset, and refining alignment tuning methods to produce safer and more reliable Korean medical LLMs.

Objectives: Developing large language models (LLMs) in biomedicine requires access to high-quality training and alignment tuning datasets. However, publicly available Korean medical preference datasets are scarce, hindering the advancement of Korean medical LLMs. This study constructs and evaluates the efficacy of the Korean Medical Preference Dataset (KoMeP), an alignment tuning dataset constructed with an automated pipeline, minimizing the high costs of human annotation. Methods: KoMeP was generated using the DAHL score, an automated hallucination evaluation metric. Five LLMs (Dolly-v2-3B, MPT-7B, GPT-4o, Qwen-2-7B, Llama-3-8B) produced responses to 8,573 biomedical examination questions, from which 5,551 preference pairs were extracted. Each pair consisted of a “chosen” response and a “rejected” response, as determined by their DAHL scores. The dataset was evaluated when trained through two different alignment tuning methods, direct preference optimization (DPO) and odds ratio preference optimization (ORPO) respectively across five different models. The KorMedMCQA benchmark was employed to assess the effectiveness of alignment tuning. Results: Models trained with DPO consistently improved KorMedMCQA performance; notably, Llama-3.1-8B showed a 43.96% increase. In contrast, ORPO training produced inconsistent results. Additionally, English-to-Korean transfer learning proved effective, particularly for English-centric models like Gemma-2, whereas Korean-to-English transfer learning achieved limited success. Instruction tuning with KoMeP yielded mixed outcomes, which suggests challenges in dataset formatting. Conclusions KoMeP is the first publicly available Korean medical preference dataset and significantly improves alignment tuning performance in LLMs. The DPO method outperforms ORPO in alignment tuning. Future work should focus on expanding KoMeP, developing a Korean-native dataset, and refining alignment tuning methods to produce safer and more reliable Korean medical LLMs.

Thelazia callipaeda (Spirurida, Thelaziidae), a parasitic eye worm causing thelaziasis has been reported in humans and dogs in Korea. However, its occurrence in other potential reservoir hosts, including wild animals, remains unclear. In the present study, we described the 2 cases of thelaziasis from both of feral and captive wild animal in Korea. In August and November 2017, 2 cases of the parasitic infections were found in the third eyelid of rescued leopard cat Prionailurus bengalensis and reared raccoon dog Nyctereutes procyonoides at the Chungnam Wild Animal Rescue Center. A total of 20 and 24 worms were detected from the left and right eyes of leopard cat, respectively. In the left eye of the raccoon dog, 5 worms were recognized. Male worms were 969–11,860 μm long (10,600 μm on average) and 300–320 μm width (315 μm on average). Female worms were 13,430–15,330 (14,480) μm long and 320–370 (344) μm wide in size. They commonly had a characteristic scalariform buccal cavity and short esophagus. The vulva openings were located at the anterior of esophago-intestinal junction in females. The thelaziasis is reported in 2 species of wildlife, P. bengalensis and N. procyonoides, for the first time in Korea.