Hugging Face's logo

Datasets:
HuggingFaceBio
/
perturbation-bench

Tasks:
Text Generation
Modalities:
Tabular
Text
Formats:
parquet
Size:
10K - 100K
Tags:
biology
genomics
dna
benchmark
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
mauryzio commited on
Commit
9839891
·
verified ·
1 Parent(s): c522916

Remove script_version from dataset_info schema

Browse files
Files changed (1) hide show
  1. README.md +249 -2
README.md CHANGED
@@ -20,6 +20,253 @@ configs:
20
  data_files:
21
  - split: test
22
  path: syn_mouse/test-*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
23
  ---
24
 
25
  # Carbon Perturbation Bench
@@ -37,7 +284,7 @@ altered — and asks whether the model assigns higher log-likelihood to the orig
37
  ### `syn_human` · `syn_mouse` — Synonymous codon substitution (20,000 pairs each)
38
 
39
  Codons within a real CDS are replaced with the highest-frequency synonym for the target
40
- species, while the upstream and
41
  downstream flanking sequence is left unchanged. Amino acid identity is preserved by
42
  construction. The model should prefer the natural codon usage over the artificially
43
  optimised variant.
@@ -77,7 +324,7 @@ All configs share the key columns:
77
  | `benchmark_id` | Unique row identifier |
78
 
79
  Config-specific columns (e.g. `patch_start_in_seq`, `n_codons_changed`,
80
- `fraction_codons_changed`, `codon_usage_source`) are described in `dataset_info` above.
81
 
82
  ---
83
 
 
20
  data_files:
21
  - split: test
22
  path: syn_mouse/test-*
23
+ dataset_info:
24
+ - config_name: syn_human
25
+ features:
26
+ - name: benchmark_id
27
+ dtype: string
28
+ - name: annotation_source
29
+ dtype: string
30
+ - name: genome_assembly
31
+ dtype: string
32
+ - name: codon_usage_source
33
+ dtype: string
34
+ - name: recoding_mode
35
+ dtype: string
36
+ - name: optimization_rate
37
+ dtype: float64
38
+ - name: random_seed
39
+ dtype: int64
40
+ - name: window_size_bp
41
+ dtype: int64
42
+ - name: species
43
+ dtype: string
44
+ - name: chr
45
+ dtype: string
46
+ - name: chrom_length
47
+ dtype: int64
48
+ - name: strand
49
+ dtype: string
50
+ - name: cds_genome_start
51
+ dtype: int64
52
+ - name: cds_genome_end
53
+ dtype: int64
54
+ - name: window_genome_start
55
+ dtype: int64
56
+ - name: window_genome_end
57
+ dtype: int64
58
+ - name: upstream_actual
59
+ dtype: int64
60
+ - name: downstream_actual
61
+ dtype: int64
62
+ - name: is_clamped
63
+ dtype: int64
64
+ - name: transcript_id
65
+ dtype: string
66
+ - name: gene_id
67
+ dtype: string
68
+ - name: gene_name
69
+ dtype: string
70
+ - name: exon_rank
71
+ dtype: int64
72
+ - name: is_first_exon
73
+ dtype: int64
74
+ - name: frame
75
+ dtype: int64
76
+ - name: phase_gtf
77
+ dtype: int64
78
+ - name: cds_start_in_seq
79
+ dtype: int64
80
+ - name: cds_end_in_seq
81
+ dtype: int64
82
+ - name: cds_length
83
+ dtype: int64
84
+ - name: patch_start_in_seq
85
+ dtype: int64
86
+ - name: patch_end_in_seq
87
+ dtype: int64
88
+ - name: patch_length
89
+ dtype: int64
90
+ - name: n_codons_total
91
+ dtype: int64
92
+ - name: n_codons_eligible
93
+ dtype: int64
94
+ - name: n_codons_changed
95
+ dtype: int64
96
+ - name: fraction_codons_changed
97
+ dtype: float64
98
+ - name: original_sequence
99
+ dtype: string
100
+ - name: sequence
101
+ dtype: string
102
+ splits:
103
+ - name: test
104
+ num_bytes: 335904126
105
+ num_examples: 20000
106
+ download_size: 152737063
107
+ dataset_size: 335904126
108
+ - config_name: syn_mouse
109
+ features:
110
+ - name: benchmark_id
111
+ dtype: string
112
+ - name: annotation_source
113
+ dtype: string
114
+ - name: genome_assembly
115
+ dtype: string
116
+ - name: codon_usage_source
117
+ dtype: string
118
+ - name: recoding_mode
119
+ dtype: string
120
+ - name: optimization_rate
121
+ dtype: float64
122
+ - name: random_seed
123
+ dtype: int64
124
+ - name: window_size_bp
125
+ dtype: int64
126
+ - name: species
127
+ dtype: string
128
+ - name: chr
129
+ dtype: string
130
+ - name: chrom_length
131
+ dtype: int64
132
+ - name: strand
133
+ dtype: string
134
+ - name: cds_genome_start
135
+ dtype: int64
136
+ - name: cds_genome_end
137
+ dtype: int64
138
+ - name: window_genome_start
139
+ dtype: int64
140
+ - name: window_genome_end
141
+ dtype: int64
142
+ - name: upstream_actual
143
+ dtype: int64
144
+ - name: downstream_actual
145
+ dtype: int64
146
+ - name: is_clamped
147
+ dtype: int64
148
+ - name: transcript_id
149
+ dtype: string
150
+ - name: gene_id
151
+ dtype: string
152
+ - name: gene_name
153
+ dtype: string
154
+ - name: exon_rank
155
+ dtype: int64
156
+ - name: is_first_exon
157
+ dtype: int64
158
+ - name: frame
159
+ dtype: int64
160
+ - name: phase_gtf
161
+ dtype: int64
162
+ - name: cds_start_in_seq
163
+ dtype: int64
164
+ - name: cds_end_in_seq
165
+ dtype: int64
166
+ - name: cds_length
167
+ dtype: int64
168
+ - name: patch_start_in_seq
169
+ dtype: int64
170
+ - name: patch_end_in_seq
171
+ dtype: int64
172
+ - name: patch_length
173
+ dtype: int64
174
+ - name: n_codons_total
175
+ dtype: int64
176
+ - name: n_codons_eligible
177
+ dtype: int64
178
+ - name: n_codons_changed
179
+ dtype: int64
180
+ - name: fraction_codons_changed
181
+ dtype: float64
182
+ - name: original_sequence
183
+ dtype: string
184
+ - name: sequence
185
+ dtype: string
186
+ splits:
187
+ - name: test
188
+ num_bytes: 336064778
189
+ num_examples: 20000
190
+ download_size: 152659316
191
+ dataset_size: 336064778
192
+ - config_name: motif_human
193
+ features:
194
+ - name: benchmark_id
195
+ dtype: string
196
+ - name: annotation_source
197
+ dtype: string
198
+ - name: genome_assembly
199
+ dtype: string
200
+ - name: upstream_flank_bp
201
+ dtype: int64
202
+ - name: motif
203
+ dtype: string
204
+ - name: actual_motif
205
+ dtype: string
206
+ - name: patch_len_bp
207
+ dtype: int64
208
+ - name: start_after_bp
209
+ dtype: int64
210
+ - name: random_seed
211
+ dtype: int64
212
+ - name: species
213
+ dtype: string
214
+ - name: chr
215
+ dtype: string
216
+ - name: chrom_length
217
+ dtype: int64
218
+ - name: strand
219
+ dtype: string
220
+ - name: cds_genome_start
221
+ dtype: int64
222
+ - name: cds_genome_end
223
+ dtype: int64
224
+ - name: window_genome_start
225
+ dtype: int64
226
+ - name: window_genome_end
227
+ dtype: int64
228
+ - name: upstream_actual
229
+ dtype: int64
230
+ - name: is_clamped
231
+ dtype: int64
232
+ - name: transcript_id
233
+ dtype: string
234
+ - name: gene_id
235
+ dtype: string
236
+ - name: gene_name
237
+ dtype: string
238
+ - name: exon_rank
239
+ dtype: int64
240
+ - name: is_first_exon
241
+ dtype: int64
242
+ - name: frame
243
+ dtype: int64
244
+ - name: phase_gtf
245
+ dtype: int64
246
+ - name: cds_start_in_seq
247
+ dtype: int64
248
+ - name: cds_end_in_seq
249
+ dtype: int64
250
+ - name: cds_length
251
+ dtype: int64
252
+ - name: patch_start_in_cds
253
+ dtype: int64
254
+ - name: patch_end_in_cds
255
+ dtype: int64
256
+ - name: patch_start_in_seq
257
+ dtype: int64
258
+ - name: patch_end_in_seq
259
+ dtype: int64
260
+ - name: original_sequence
261
+ dtype: string
262
+ - name: sequence
263
+ dtype: string
264
+ splits:
265
+ - name: test
266
+ num_bytes: 334784354
267
+ num_examples: 20000
268
+ download_size: 152132105
269
+ dataset_size: 334784354
270
  ---
271
 
272
  # Carbon Perturbation Bench
 
284
  ### `syn_human` · `syn_mouse` — Synonymous codon substitution (20,000 pairs each)
285
 
286
  Codons within a real CDS are replaced with the highest-frequency synonym for the target
287
+ species (`recoding_mode = "optimize"`, `optimization_rate = 1.0`), while the upstream and
288
  downstream flanking sequence is left unchanged. Amino acid identity is preserved by
289
  construction. The model should prefer the natural codon usage over the artificially
290
  optimised variant.
 
324
  | `benchmark_id` | Unique row identifier |
325
 
326
  Config-specific columns (e.g. `patch_start_in_seq`, `n_codons_changed`,
327
+ `fraction_codons_changed`, `codon_usage_source`) are described in the dataset info above.
328
 
329
  ---
330