A Multi-Criteria Ranking Pipeline for Somatic Variant Candidate Prioritization in Cardiac Whole-Exome Sequencing
Provisional site (2026-10-07). The framework is designed; components and prototypes are in development and validation is ongoing. No performance results are reported yet.
Abstract
Whole-exome sequencing (WES) of cardiac tissue without matched normal samples yields candidate variants whose technical support and biological relevance require separate review. Multiple callers may produce large, discordant candidate lists containing sequencing or alignment artifacts, creating a need for explicit prioritization and traceable evidence. This thesis develops a multi-criteria ranking pipeline that supports, rather than replaces, expert review. Candidates from the union of several somatic callers are ranked solely on technical evidence by a learned scoring model, with comparison against simple baselines planned on identical candidates and labels. A fixed Top-K shortlist is then annotated in a separate stage with functional impact, expression, multiome context, conflicting evidence, uncertainty and provenance, so that biological information informs review without silently influencing selection. The output is a traceable evidence report for each shortlisted candidate. A grounded language-model layer is being prototyped to summarize and audit this evidence; it cannot change rankings or assert biological truth. The main contribution is an integrated, traceable review workflow with an explicit evaluation protocol for technical scoring.
Keywords
somatic variant prioritization · cardiac whole-exome sequencing · multi-criteria ranking · learned technical scoring · expert review · evidence traceability
On this site
- Motivation: the review problem and the scope of the work
- Framework: the pipeline, from technical ranking to grounded review
- Evaluation: metrics, baselines and label policy
- Progress: what is designed, implemented, prototyped or ongoing
- Resources: official links and release status
한국어 요약
본 연구는 심장 조직 전장 엑솜 시퀀싱(WES)에서 여러 체세포 변이 검출기의 후보를 합친 뒤, 기술적 근거만으로 학습된 순위 모델이 후보를 정렬하고, 고정된 상위 K개 후보에 대해 기능 영향·발현·멀티오믹스 맥락·상충 근거·불확실성·출처를 별도 단계에서 정리하여 전문가 검토용 근거 보고서를 만드는 파이프라인을 다룹니다.
목표는 체세포 변이 진단이 아니라 전문가 검토 지원입니다. 언어 모델은 근거 요약·감사용 프로토타입이며 순위를 바꾸거나 사실을 판정하지 않습니다. 전체 설계는 정의되었으나 모델 구조는 아직 확정 전이며, 구성 요소 통합과 검증이 진행 중입니다. 최종 웹사이트는 결과 제출 기한(2026-12-31)까지 완성할 예정입니다.
Research information
Author: 곽자현 · Advisor: 송길태 · Pusan National University
Expected degree: February 2027 · kwjh1004@pusan.ac.kr