Most substitutions need more than one base change and an exact test tells which deposited scans measure the route

Design methods that reach a target through an intermediate take two substitutions at one position, and no deep mutational scan built from a single parental sequence reports the second. The detour is common, since 230 of the 380 ordered pairs of amino acids are more than one base change apart. Both substitutions are nevertheless present in the public record wherever one protein has been scanned on two parental backgrounds that differ at a position. We assemble every such comparison in MaveDB, 102,421 complete two-step routes over 65 protein clusters. The difference between a route and its two legs is a double difference, which is the epistasis between the tested position and the other positions at which the two backgrounds differ. Where the backgrounds differ nowhere else, that quantity is zero by construction. The rank correlation over the target residues must then return one, on any scale and against no null, which makes the comparison an exact test. Over the 741 qualifying pairs across 46 proteins the median is 0.965, and 34 of the 46 proteins return a median at or above 0.9. The arithmetic therefore holds across the population the record supplies, short natural domains drawn from families of near-identical sequences. Which pairs fail is not written in the record, since pairs drawn from one experiment set return 0.34 and pairs from unrelated submissions 0.97. The null such comparisons are read against fails the same test, returning +0.227 where it must return zero, more than the shift the comparison reports. Matching the control positions on the dispersion of the effects ranked halves that bias to +0.104. Two deposits of one protein agree at 0.642 where both score the same position. That bounds any predictor tested against one deposit at 0.80 for a position drawn at random, and at 0.90 and 0.98 where positions are scored more reliably. The test costs one correlation and settles, before any analysis, whether two deposits can be compared.

Authors

Institutions

Publication Details

Journal
bioRxiv (Cold Spring Harbor Laboratory)
Published
2026-11-18
DOI
https://doi.org/10.64898/2026.11.17.767252
Primary Topic
Evolution and Genetic Dynamics
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Most substitutions need more than one base change and an exact test tells which deposited scans measure the route

Taner Karagöl, Alper Karagöl
3 links
bioRxiv (Cold Spring Harbor Laboratory)
Evolution and Genetic Dynamics
preprint

Most substitutions need more than one base change and an exact test tells which deposited scans measure the route

Taner Karagöl, Alper Karagöl
preprint en

Abstract

Design methods that reach a target through an intermediate take two substitutions at one position, and no deep mutational scan built from a single parental sequence reports the second. The detour is common, since 230 of the 380 ordered pairs of amino acids are more than one base change apart. Both substitutions are nevertheless present in the public record wherever one protein has been scanned on two parental backgrounds that differ at a position. We assemble every such comparison in MaveDB, 102,421 complete two-step routes over 65 protein clusters. The difference between a route and its two legs is a double difference, which is the epistasis between the tested position and the other positions at which the two backgrounds differ. Where the backgrounds differ nowhere else, that quantity is zero by construction. The rank correlation over the target residues must then return one, on any scale and against no null, which makes the comparison an exact test. Over the 741 qualifying pairs across 46 proteins the median is 0.965, and 34 of the 46 proteins return a median at or above 0.9. The arithmetic therefore holds across the population the record supplies, short natural domains drawn from families of near-identical sequences. Which pairs fail is not written in the record, since pairs drawn from one experiment set return 0.34 and pairs from unrelated submissions 0.97. The null such comparisons are read against fails the same test, returning +0.227 where it must return zero, more than the shift the comparison reports. Matching the control positions on the dispersion of the effects ranked halves that bias to +0.104. Two deposits of one protein agree at 0.642 where both score the same position. That bounds any predictor tested against one deposit at 0.80 for a position drawn at random, and at 0.90 and 0.98 where positions are scored more reliably. The test costs one correlation and settles, before any analysis, whether two deposits can be compared.

bioRxiv (Cold Spring Harbor Laboratory)
Massachusetts Institute of Technology (US)
Evolution and Genetic Dynamics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon