BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260625T133339Z
LOCATION:Bldg. 6 - 001 - Plenary Room
DTSTART;TZID=Europe/Stockholm:20260630T123200
DTEND;TZID=Europe/Stockholm:20260630T123300
UID:submissions.pasc-conference.org_PASC26_sess129_posC101@linklings.com
SUMMARY:ACMP04 - Fine-Tuning Large Language Models for HPO-Term Recognitio
 n
DESCRIPTION:Srinithi Krishnamoorthy (Cornell University)\n\nLarge language
  models (LLMs) offer a promising approach for extracting structured medica
 l information from free-text clinical notes. This work investigates fine-t
 uning LLMs for Human Phenotype Ontology (HPO) term recognition, a core tas
 k in clinical genetics that requires accurate identification and normaliza
 tion of phenotypic abnormalities. Multiple model architectures, including 
 LLaMA2 and Falcon models at different parameter scales, are fine-tuned on 
 HPC infrastructure utilizing NVIDIA A100 GPUs using annotated clinical dat
 a, with and without supplementary HPO ontology knowledge, and several extr
 action strategies are evaluated. The results show that LLaMA2 7B consisten
 tly outperforms larger and alternative models, indicating that increased m
 odel size does not necessarily improve performance in data-constrained cli
 nical settings. Incorporating structured HPO knowledge improves training s
 tability and generalization, while a joint “Spans and Terms” extraction ap
 proach yields the highest accuracy. The best-performing configuration achi
 eves an F1 score of 0.6942 under exact matching and 0.7375 with fuzzy matc
 hing. These findings highlight the importance of model choice and extracti
 on design for reliable phenotype normalization from clinical text.\n\nSess
 ion Chair: Tobias Hodel (University of Bern, Switzerland)\n\n
END:VEVENT
END:VCALENDAR