Resolving ambiguous non-coding screen hits.
Nearest-gene heuristics misassign ~46% of non-coding hits (measured on 72 held-out CRISPRi pairs). TrueLocus nominates causal distal targets using 3D contact priors and multi-promoter competition without requiring cell-type Hi-C.
Held-Out Empirical Benchmark
Evaluated on 72 experimental CRISPRi enhancer-gene pairs in K562 cells (Schraivogel et al. 2020, Nature Methods). Code was frozen and committed before evaluation.
| Method | Top-1 Accuracy | Top-3 Target Recall | Input Required |
|---|---|---|---|
| Nearest-TSS Baseline | 54.2% (39/72) | 73.6% (53/72) | 1D Distance |
| TrueLocus Engine | 69.4% (50/72) | 84.7% (61/72) | Sequence + CAGE (No Hi-C) |
| Broad ABC (Nasser 2021) | 75.0% (54/72) | 94.4% (68/72) | Full K562 Hi-C + ChIP + ATAC |
Paired exact test: TrueLocus vs Nearest Top-1 p=0.007 (McNemar exact, b=13, c=2). On the 33 non-nearest loci where enhancers skip the closest promoter, nearest-gene Top-3 recall collapses to 42%, while TrueLocus recovers 70%.
Per-locus results tables and full method description available on request; code review under NDA for qualified evaluators.
Services & Engagements
Inquiries & Contact
TrueLocus operates directly with functional genomics and target discovery screening teams. To test the engine on your hits or request full per-locus benchmark logs: