Update data splits and README

#5
by arvban - opened
Files changed (2) hide show
  1. README.md +6 -8
  2. metadata.csv +2 -2
README.md CHANGED
@@ -79,12 +79,12 @@ The dataset provides predefined train, validation, and test partitions. Complexe
79
 
80
  | Split | Complexes | Share |
81
  |---|---:|---:|
82
- | train | 131,936 | 43.10% |
83
- | validation | 53,548 | 17.49% |
84
- | test | 52,132 | 17.03% |
85
- | deleaked | 68,521 | 22.38% |
86
  | total | 306,137 | 100.00% |
87
- | usable benchmark | 237,616 | 77.62% |
88
 
89
  The usable benchmark consists of the train, validation, and test partitions only.
90
 
@@ -96,8 +96,6 @@ The split was constructed using an interface-level procedure adapted from [PINDE
96
 
97
  The resulting Foldseek interface-similarity graph was thresholded using the PINDER clustering threshold of 0.70, and weighted asynchronous label propagation was used to assign chain-interface cluster labels. Each complex was assigned to a paired-interface cluster defined by its target-side and binder-side interface cluster identifiers. Train, validation, and test assignment was performed at the paired-interface-cluster level.
98
 
99
- After holdout clusters were selected, PINDER’s depth-2 transitive deleaking logic was applied using the deleaking threshold of 0.55. Candidate training complexes connected to validation or test complexes through one or two interface-similarity graph edges were assigned to the `deleaked` partition and excluded from the usable benchmark.
100
-
101
  After holdout clusters were selected, PINDER’s depth-2 transitive deleaking procedure was applied. Candidate training complexes connected to validation or test complexes through one or two edges in either the Foldseek interface-similarity graph or a separately constructed interface-filtered MMseqs sequence-similarity graph were assigned to the deleaked partition and excluded from the usable benchmark.
102
 
103
  ## Redundancy Audit
@@ -106,7 +104,7 @@ As an independent audit, we assessed whether ProteinMPNN sister variants derived
106
 
107
  Among 120,609 backbones with ProteinMPNN sister variants, 120,082 were contained within a single paired-interface cluster, whereas 527 were split across multiple clusters. 99.56% of multi-member backbone groups were captured by the paired-interface clustering procedure.
108
 
109
- After deleaking, only 14 of the 85,036 multi-member backbone groups in the usable benchmark (0.016%) have sister variants split between `train` and the held-out (`validation` or `test`) split.
110
 
111
 
112
  ## Intended Use
 
79
 
80
  | Split | Complexes | Share |
81
  |---|---:|---:|
82
+ | train | 205,553 | 67.14% |
83
+ | validation | 21,205 | 6.93% |
84
+ | test | 21,955 | 7.17% |
85
+ | deleaked | 57,424 | 18.76% |
86
  | total | 306,137 | 100.00% |
87
+ | usable benchmark | 248,713 | 81.24% |
88
 
89
  The usable benchmark consists of the train, validation, and test partitions only.
90
 
 
96
 
97
  The resulting Foldseek interface-similarity graph was thresholded using the PINDER clustering threshold of 0.70, and weighted asynchronous label propagation was used to assign chain-interface cluster labels. Each complex was assigned to a paired-interface cluster defined by its target-side and binder-side interface cluster identifiers. Train, validation, and test assignment was performed at the paired-interface-cluster level.
98
 
 
 
99
  After holdout clusters were selected, PINDER’s depth-2 transitive deleaking procedure was applied. Candidate training complexes connected to validation or test complexes through one or two edges in either the Foldseek interface-similarity graph or a separately constructed interface-filtered MMseqs sequence-similarity graph were assigned to the deleaked partition and excluded from the usable benchmark.
100
 
101
  ## Redundancy Audit
 
104
 
105
  Among 120,609 backbones with ProteinMPNN sister variants, 120,082 were contained within a single paired-interface cluster, whereas 527 were split across multiple clusters. 99.56% of multi-member backbone groups were captured by the paired-interface clustering procedure.
106
 
107
+ After deleaking, only 23 of the 88,720 multi-member backbone groups in the usable benchmark (0.026%) have sister variants split between `train` and the held-out (`validation` or `test`) split.
108
 
109
 
110
  ## Intended Use
metadata.csv CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:91a16e63009f562a5b567e9a35636ff991a42ba91802b9c76da4a18d849d8378
3
- size 160780416
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:396499c0faff6f731c8ef0ed76c6bda02f85ce7a59e4d9eb919f8a10a986cb15
3
+ size 160137408