Dataset analysis: `fc_retraining_v1`

Run id 20260517-202905_data_fc_retraining_v1_seed0 · 2026-05-18T12:34:57+00:00 · championsnet

TL;DR

1,425,757 rows / 1,425,757 unique sequences (0 duplicates).
Mean labels per row: 1.37 / 8. Rows with all 8 labels: 4,181. Rows with zero labels: 5,227.
WT length rate: 100.00%; canonical AA rate: 100.00%; N-glyco motif rate: 14.37%.
Mean edit distance to WT: 5.28 (median 5, p99 15).
Overall per-position entropy: 0.18 bits; 13/232 positions above 1 bit.

Basic

rows	1,425,757
columns	80
unique_seq_hash	1,425,757
duplicate_rows	0

source_dataset distribution

source	rows	fraction
`dms`	317,482	22.27%
`dms+paper_refiltered`	6,610	0.46%
`paper_refiltered`	1,101,665	77.27%

label_source distribution

label_source	rows	fraction
`merged`	5,060	0.35%
`null`	5,227	0.37%
`sort_label`	1,415,470	99.28%

Label coverage

Labels per row

labels per row	rows	fraction	distribution
0	5,227	0.37%	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
1	1,101,278	77.24%	████████████████████████████████
2	203,888	14.30%	█████░░░░░░░░░░░░░░░░░░░░░░░░░░░
3	64,643	4.53%	█░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
4	25,889	1.82%	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
5	11,456	0.80%	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
6	5,532	0.39%	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
7	3,663	0.26%	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
8	4,181	0.29%	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░

Per-receptor balance

Receptor	binders	non-binders	missing	missing %	bind % (of called)
`R1`	120,377	188,883	1,116,497	78.31%	38.92%
`FcRn6`	99,628	192,117	1,134,012	79.54%	34.15%
`R2AH`	25,899	126,341	1,273,517	89.32%	17.01%
`R2AR`	45,477	181,946	1,198,334	84.05%	20.00%
`R2B`	54,463	151,446	1,219,848	85.56%	26.45%
`R3AF`	21,705	189,190	1,214,862	85.21%	10.29%
`R3AV`	27,213	189,669	1,208,875	84.79%	12.55%
`R3B`	34,156	307,590	1,084,011	76.03%	9.99%

Profile strings

2,319 unique profile strings; 4,181 rows (0.29%) have all 8 receptors labelled.

Top 20 most common profiles

profile	rows	fraction
`-------0`	190,968	13.39%
`0-------`	120,979	8.49%
`-0------`	120,004	8.42%
`------0-`	113,115	7.93%
`-----0--`	109,389	7.67%
`---0----`	108,776	7.63%
`----0---`	84,071	5.90%
`--0-----`	69,748	4.89%
`1-------`	54,040	3.79%
`-1------`	52,947	3.71%
`----1---`	20,791	1.46%
`---1----`	18,336	1.29%
`-------1`	13,535	0.95%
`0------0`	9,407	0.66%
`-0-----0`	9,252	0.65%
`--1-----`	8,796	0.62%
`------1-`	8,692	0.61%
`-----0-0`	7,265	0.51%
`-----1--`	7,091	0.50%
`-----00-`	6,835	0.48%

Sequence features

wt_length_rate	1.00
canonical_aa_rate	1.00
n_glyco_motif_rate	0.1437
in_variable_region_only_rate	0.8956

Edit distance to WT

mean = 5.28, std = 2.63, median = 5, p90 = 8, p99 = 15, max = 61.

edit distance	rows	histogram
`0`	1	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
`1`	3,804	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
`2`	253,735	█████████░░░░░░░░░░░░░░░░░░░░░
`3-4`	209,448	███████░░░░░░░░░░░░░░░░░░░░░░░
`5-7`	810,754	██████████████████████████████
`8-9`	73,969	██░░░░░░░░░░░░░░░░░░░░░░░░░░░░
`10-14`	58,538	██░░░░░░░░░░░░░░░░░░░░░░░░░░░░
`15-19`	11,366	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
`20-29`	3,987	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
`30-49`	135	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
`>=50`	20	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░

Mean edit distance, by per-receptor label

Receptor	binders	non-binders	missing
`R1`	mean=4.90, median=6 (n=120,377)	mean=5.33, median=5 (n=188,883)	mean=5.31, median=5 (n=1,116,497)
`FcRn6`	mean=4.41, median=5 (n=99,628)	mean=5.64, median=6 (n=192,117)	mean=5.29, median=5 (n=1,134,012)
`R2AH`	mean=2.79, median=2 (n=25,899)	mean=5.19, median=5 (n=126,341)	mean=5.33, median=5 (n=1,273,517)
`R2AR`	mean=2.79, median=2 (n=45,477)	mean=5.31, median=5 (n=181,946)	mean=5.37, median=5 (n=1,198,334)
`R2B`	mean=2.37, median=2 (n=54,463)	mean=4.78, median=5 (n=151,446)	mean=5.47, median=6 (n=1,219,848)
`R3AF`	mean=2.69, median=2 (n=21,705)	mean=5.67, median=6 (n=189,190)	mean=5.26, median=5 (n=1,214,862)
`R3AV`	mean=3.14, median=2 (n=27,213)	mean=5.73, median=6 (n=189,669)	mean=5.25, median=5 (n=1,208,875)
`R3B`	mean=2.51, median=2 (n=34,156)	mean=4.91, median=5 (n=307,590)	mean=5.46, median=6 (n=1,084,011)

Read counts per receptor

Receptor	n with count	mean	median	p90	p99	max
`R1`	309,260	16.50	3	8	87	1,409,172
`FcRn6`	291,745	17.14	3	8	87	1,245,548
`R2AH`	152,240	10.25	3	7	103	296,845
`R2AR`	227,423	12.11	3	7	112	601,641
`R2B`	205,909	20.95	3	8	174	996,987
`R3AF`	210,895	6.60	3	7	45	106,112
`R3AV`	216,882	6.68	3	7	43	135,940
`R3B`	341,746	16.36	3	8	77	1,718,201

Sequence diversity (per-position Shannon entropy)

Overall (sampled): n=200,000, mean entropy = 0.18 bits, max position = 2.49, positions above 1 bit = 13/232, zero-entropy positions = 7.

Sequences with all 8 labels: n=4,181, mean entropy = 0.08 bits, max position = 2.12, positions above 1 bit = 4/232, zero-entropy positions = 7.

Per receptor (binders vs non-binders)

Receptor	binders	non-binders
`R1`	n=100,000, mean H = 0.12	n=100,000, mean H = 0.17
`FcRn6`	n=99,628, mean H = 0.13	n=100,000, mean H = 0.16
`R2AH`	n=25,899, mean H = 0.11	n=100,000, mean H = 0.17
`R2AR`	n=45,477, mean H = 0.11	n=100,000, mean H = 0.17
`R2B`	n=54,463, mean H = 0.10	n=100,000, mean H = 0.16
`R3AF`	n=21,705, mean H = 0.11	n=100,000, mean H = 0.17
`R3AV`	n=27,213, mean H = 0.11	n=100,000, mean H = 0.17
`R3B`	n=34,156, mean H = 0.10	n=100,000, mean H = 0.16

Per-source breakdown

`dms`

317,482 rows, mean labels/row = 1.49

Receptor	binders	non-binders	missing	missing %	bind % (of called)
`R1`	21,379	39,570	256,533	80.80%	35.08%
`FcRn6`	21,042	28,811	267,629	84.30%	42.21%
`R2AH`	13,832	25,174	278,476	87.71%	35.46%
`R2AR`	28,344	37,473	251,665	79.27%	43.06%
`R2B`	43,092	50,766	223,624	70.44%	45.91%
`R3AF`	11,535	16,817	289,130	91.07%	40.68%
`R3AV`	11,089	14,629	291,764	91.90%	43.12%
`R3B`	24,987	83,893	208,602	65.71%	22.95%

`dms+paper_refiltered`

6,610 rows, mean labels/row = 4.14

Receptor	binders	non-binders	missing	missing %	bind % (of called)
`R1`	1,552	1,862	3,196	48.35%	45.46%
`FcRn6`	2,638	782	3,190	48.26%	77.13%
`R2AH`	1,640	1,287	3,683	55.72%	56.03%
`R2AR`	1,987	1,270	3,353	50.73%	61.01%
`R2B`	2,331	1,821	2,458	37.19%	56.14%
`R3AF`	1,530	1,355	3,725	56.35%	53.03%
`R3AV`	1,718	1,260	3,632	54.95%	57.69%
`R3B`	1,707	2,654	2,249	34.02%	39.14%

`paper_refiltered`

1,101,665 rows, mean labels/row = 1.32

Receptor	binders	non-binders	missing	missing %	bind % (of called)
`R1`	97,446	147,451	856,768	77.77%	39.79%
`FcRn6`	75,948	162,524	863,193	78.35%	31.85%
`R2AH`	10,427	99,880	991,358	89.99%	9.45%
`R2AR`	15,146	143,203	943,316	85.63%	9.56%
`R2B`	9,040	98,859	993,766	90.21%	8.38%
`R3AF`	8,640	171,018	922,007	83.69%	4.81%
`R3AV`	14,406	173,780	913,479	82.92%	7.66%
`R3B`	7,462	221,043	873,160	79.26%	3.27%

Generated by scripts/data/analyze_dataset.py. Machine-readable companion: analysis.yaml in the same run directory.

Dataset analysis: fc_retraining_v1