Benchmarks

This section summarizes the benchmark evidence used by the zsasa manuscript. The current headline results come from the pinned zsasa v0.6.0 benchmark harness in zsasa-benchmarks, with figures copied from the paper and benchmark repositories.

The older exploratory benchmark pages have been replaced with the current benchmark suite layout. The only pre-pinned result retained here is the SwissProt-scale batch benchmark, labeled as legacy context on the Batch page.

Benchmark suites

Suite	Purpose	Primary dataset	Main readouts
Validation	Agreement with established SASA implementations	E. coli AFDB and 5wvo_C trajectory	R², relative error
Batch throughput	Proteome-scale directory processing	E. coli and Human AFDB	Runtime, structures/s, RSS, scaling
Single-file stress tests	Large structures and parser-heavy cases	8 curated structures up to 4.5M atoms	Runtime, RSS, parse/SASA timing
MD trajectories	Streaming frame-wise trajectory SASA	5wvo_C, 6sup_A, 5vz0_A	Frames/s, RSS, comparator speedup

Test environment

The pinned current pinned benchmarks were run on one consumer laptop:

Item	Value
Machine	MacBook Pro (Mac16,1)
Chip	Apple M4
Cores	10 total: 4 performance + 6 efficiency
Memory	32 GB
OS	macOS 26.2
Timing	hyperfine 1.20.0, with warmups and measured runs per suite
Tool pinning	Nix for native tools, uv for Python dependencies
zsasa version	v0.6.0

Interpreting the numbers

• Speedups are comparator runtime divided by zsasa runtime; higher is better.
• RSS is peak resident set size; lower memory means a higher RSS-reduction ratio.
• Exact f64/f32 modes target numerical continuity with matched Shrake--Rupley outputs.
• Bitmask mode is a throughput-oriented approximation with an explicit validation envelope.
• Batch FreeSASA timings use a freesasa_batch wrapper because upstream FreeSASA has no native directory mode.

Evidence sources

• Manuscript repository: N283T/zsasa-paper
• Benchmark harness and result tables: N283T/zsasa-benchmarks
• Feature comparison: Comparison with Other Tools