acceptodds
Desk rejected at ICLR 2027

LocusGate: Operator-Neutral Graph Workspaces for Composable Pangenome Queries

Desk rejected at ICLR 2027, so left out of the paper lists and the map.

Abstract

Multi-stage pangenome analyses revisit one locus for sequence membership, path extraction, and neighborhood inspection, yet independently localized stages can expose different graph evidence. We make locality an explicit operator-neutral value with LocusGate, which materializes before dispatch from complete path-window anchors and deterministic budgeted expansion. The certificate binds visible graph content, construction settings, canonical ordering, and the answer domain of each consumer. This interface separates three guarantees: exact membership over declared embedded paths for , deterministic extraction complete within , and bounded neighborhood inspection over the certified traversal domain. We prove conditioned exactness for membership and show that certificate equality makes handoff and deterministic rematerialization produce identical normalized outputs for a fixed deterministic operator. By making region identity, guarantee scope, and immutable state travel together, LocusGate gives multi-stage pangenome queries one named body of graph evidence rather than operator-private locality choices.

Primary area:
other topics in machine learning (i.e., none of the above)
Keywords:
biomedical and clinical ai, LocusGate

Related papers

Discussion (0)

Sign in to comment.