Performance & Accuracy

eskaks implements the classical substitution models with precomputed lookup tables, which makes it dramatically faster than the established tools while staying numerically accurate.

Speed

Dataset	eskaks (4t)	KaKs_Calculator	PAML yn00	BioPython	Speedup
20 seq × 300 bp	2 ms	34 ms	8 ms	610 ms	17×
100 seq × 3 kb	6 ms	7,703 ms	697 ms	111,619 ms	1,280×
500 seq × 3 kb	74 ms	195,456 ms	-	-	2,641×

Output is deterministic regardless of the number of --workers threads.

Accuracy

The Li model achieves R² = 1.0 against KaKs_Calculator's LPB implementation. Full accuracy data and the benchmarking methodology are in benchmarks/.

Feature comparison

	eskaks	KaKs_Calculator	BioPython	PAML yn00
Nei-Gojobori model	✅	✅	✅	✅
Li/LPB93 model	✅	✅	❌	❌
Per-gene pN/pS from VCF	✅	❌	❌	❌
Neutrality test + FDR	✅	❌	❌	❌
Interactive HTML report	✅	❌	❌	❌
Custom genetic codes	✅ (20 tables)	❌	❌	Limited
JSON output / stdin pipe	✅	❌	❌	❌
Parallel	✅	❌	❌	❌
Speed (100 seq)	6 ms	7,703 ms	111,619 ms	697 ms