v4.5.2 — M2 single-mutant SHAPE data

Add `OK7a_M2_data.v4.5.2.csv`: mutate-and-map (M2) SHAPE data on single-nucleotide mutant libraries for all 20 Round 3 targets × 8 design methods (~100 mutants per design, 14,136 sequences total; median 98 mutants per design, range 52–100). Probed with 2A3 SHAPE (Ultima sequencing, 2025-09-04).

Unlike the M2R file, this file contains only `mutA` (no `mutB` or `rescue_factor`). Wild-type rows include new columns `M2_structure`, `M2_F1`, and `M2_F1_crossed_pair` giving the secondary structure inferred from the M2 Z-score contact map by ShapeKnots and its F1 accuracy vs. the target pseudoknot structure.

See README for full field descriptions.

v4.5.1

• 'Publication release candidate'.
• For clarity, change method name 'WT' to be 'Starting sequence'.
• Hot fix to specification of gRNAde vs. gRNAde-no3d in OK7a_M2R_data.v4.5.1.csv.

v4.5.0

• 'Publication release candidate'.
• Remove previous versions to minimize download sizes of latest releases.
• Rescore of OpenKnot experiments OpenKnotBench_data.v4.5.0.csv to give best score for 2 structures in two cases that had two slightly different target structures released at different times (target W09 in Rounds 1-2 and target P06 in Round 3).
• Mutate-map-rescue (M2R-seq) experiments updated to clarify gRNAde vs. gRNAde-no3d as OK7a_M2R_data.v4.5.0.csv. Fix column name ref_structure to be target_structure.

v4.3.0

v4.3.0 (10 December, 2025)

• Mutate-map-rescue (M2R-seq) experiments added as OK7a_M2R_data.v4.3.0.csv.
• These M2R-seq data include measurements on top designs from each tested method for Round 3. The top OpenKnot score design for each method for each target were selected amongst those with RNet_F1 greater than equal to 80. (If no designs passed this secondary structure accuracy cutoff, the top score design was still tested.).
• The constructs include the starting design and for each target base pair, the single mutants converting each pair's nucleotide to the other, and the double mutant which should lead to compensatory rescue of the secondary structure.
• For the double mutants, rescue factor is provided that gives the fraction of SHAPE RMSD between wildtype and the single mutants (added in quadrature for the two mutant) that is recovered by the double mutant. This number is 0 for no rescue (no evidence for pair) and is 1 for full rescue.

v4.2.0

v4.2.0 (30 November, 2025)

• Data on ‘round 4’ designs on 20 240-mer puzzles (collected in Eterna’ OpenKnot Round 7b lab puzzles), including puzzles like Q07 (‘Broad_bean_mottle_virus’) have been rescued and added.
• Data on struct2seq designs and codesign-RFdiff for 'round 4' have also been added.
• New columns target_structure, RNet_structure, RNet_F1, RNet_F1_crossed_pair have been added to allow filtering of designs whose SHAPE profiles match the target profiles but with the wrong secondary structures, as appears to have occurred in some 'round 3' designs, as evaluated by mutate-map-rescue experiments (to be released shortly).

OpenKnotBench v3.1.0

Initial public release of OpenKnotBench data, comparing AI and Eterna human designs through chemical mapping experiments.

This is v3.1.0.

All prior versions are also archived here; see README.md for historical release notes.