Dataset analysis: `fc_retraining_v2`

Run id 20260528-104945_data_fc_retraining_v2_seed0 · 2026-05-28T11:09:55+00:00 · championsnet

TL;DR

1,474,696 rows / 1,474,696 unique sequences (0 duplicates).
Mean labels per row: 1.61 / 8. Rows with all 8 labels: 5,066. Rows with zero labels: 1,808.
WT length rate: 100.00%; canonical AA rate: 100.00%; N-glyco motif rate: 14.20%.
Mean edit distance to WT: 5.26 (median 5, p99 15).
Overall per-position entropy: 0.18 bits; 13/232 positions above 1 bit.

Basic

rows	1,474,696
columns	80
unique_seq_hash	1,474,696
duplicate_rows	0

source_dataset distribution

source	rows	fraction
`costain`	48,939	3.32%
`dms`	316,623	21.47%
`dms+costain`	859	0.06%
`dms+paper_refiltered`	5,291	0.36%
`dms+paper_refiltered+costain`	1,319	0.09%
`paper_refiltered`	1,083,580	73.48%
`paper_refiltered+costain`	18,085	1.23%

label_source distribution

label_source	rows	fraction
`merged`	22,525	1.53%
`null`	1,808	0.12%
`sort_label`	1,450,363	98.35%

Label coverage

Labels per row

labels per row	rows	fraction	distribution
0	1,808	0.12%	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
1	989,676	67.11%	████████████████████████████████
2	265,740	18.02%	████████░░░░░░░░░░░░░░░░░░░░░░░░
3	109,977	7.46%	███░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
4	55,846	3.79%	█░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
5	27,121	1.84%	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
6	12,953	0.88%	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
7	6,509	0.44%	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
8	5,066	0.34%	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░

Per-receptor balance

Receptor	binders	non-binders	missing	missing %	bind % (of called)
`R1`	170,191	223,934	1,080,571	73.27%	43.18%
`FcRn6`	125,458	195,407	1,153,831	78.24%	39.10%
`R2AH`	42,196	144,638	1,287,862	87.33%	22.58%
`R2AR`	88,637	210,854	1,175,205	79.69%	29.60%
`R2B`	99,891	179,284	1,195,521	81.07%	35.78%
`R3AF`	38,090	233,271	1,203,335	81.60%	14.04%
`R3AV`	38,855	202,040	1,233,801	83.66%	16.13%
`R3B`	48,904	332,235	1,093,557	74.15%	12.83%

Profile strings

3,820 unique profile strings; 5,066 rows (0.34%) have all 8 receptors labelled.

Top 20 most common profiles

profile	rows	fraction
`-------0`	167,359	11.35%
`0-------`	112,549	7.63%
`-0------`	111,664	7.57%
`------0-`	109,712	7.44%
`-----0--`	106,264	7.21%
`---0----`	100,593	6.82%
`----0---`	72,590	4.92%
`--0-----`	64,537	4.38%
`1-------`	50,727	3.44%
`-1------`	47,777	3.24%
`---1----`	12,449	0.84%
`-0-----0`	10,331	0.70%
`0------0`	10,283	0.70%
`1----0--`	10,040	0.68%
`----1---`	9,698	0.66%
`-----0-0`	7,301	0.50%
`-----00-`	7,233	0.49%
`----0--0`	7,084	0.48%
`10------`	7,066	0.48%
`-------1`	7,009	0.48%

Sequence features

wt_length_rate	1.00
canonical_aa_rate	1.00
n_glyco_motif_rate	0.1420
in_variable_region_only_rate	0.8991

Edit distance to WT

mean = 5.26, std = 2.61, median = 5, p90 = 8, p99 = 15, max = 61.

edit distance	rows	histogram
`0`	1	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
`1`	3,805	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
`2`	257,537	█████████░░░░░░░░░░░░░░░░░░░░░
`3-4`	225,607	████████░░░░░░░░░░░░░░░░░░░░░░
`5-7`	838,150	██████████████████████████████
`8-9`	74,963	██░░░░░░░░░░░░░░░░░░░░░░░░░░░░
`10-14`	59,123	██░░░░░░░░░░░░░░░░░░░░░░░░░░░░
`15-19`	11,368	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
`20-29`	3,987	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
`30-49`	135	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
`>=50`	20	░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░

Mean edit distance, by per-receptor label

Receptor	binders	non-binders	missing
`R1`	mean=4.56, median=5 (n=170,191)	mean=5.03, median=5 (n=223,934)	mean=5.42, median=6 (n=1,080,571)
`FcRn6`	mean=3.93, median=4 (n=125,458)	mean=5.58, median=6 (n=195,407)	mean=5.35, median=5 (n=1,153,831)
`R2AH`	mean=2.51, median=2 (n=42,196)	mean=4.82, median=5 (n=144,638)	mean=5.40, median=6 (n=1,287,862)
`R2AR`	mean=2.90, median=2 (n=88,637)	mean=5.08, median=5 (n=210,854)	mean=5.47, median=6 (n=1,175,205)
`R2B`	mean=2.72, median=2 (n=99,891)	mean=4.59, median=5 (n=179,284)	mean=5.57, median=6 (n=1,195,521)
`R3AF`	mean=2.77, median=2 (n=38,090)	mean=5.31, median=5 (n=233,271)	mean=5.33, median=5 (n=1,203,335)
`R3AV`	mean=2.81, median=2 (n=38,855)	mean=5.52, median=6 (n=202,040)	mean=5.30, median=5 (n=1,233,801)
`R3B`	mean=2.38, median=2 (n=48,904)	mean=4.72, median=5 (n=332,235)	mean=5.55, median=6 (n=1,093,557)

Read counts per receptor

Receptor	n with count	mean	median	p90	p99	max
`R1`	394,125	15.60	3	10	88	1,409,172
`FcRn6`	320,865	15.63	3	8	75	1,245,548
`R2AH`	186,834	8.51	2	6	81	296,845
`R2AR`	299,491	12.39	2	8	92	601,641
`R2B`	279,175	22.32	2	10	137	996,987
`R3AF`	271,361	8.68	3	8	56	204,707
`R3AV`	240,895	6.08	3	7	38	135,940
`R3B`	381,139	14.76	3	8	62	1,718,201

Sequence diversity (per-position Shannon entropy)

Overall (sampled): n=200,000, mean entropy = 0.18 bits, max position = 2.48, positions above 1 bit = 13/232, zero-entropy positions = 7.

Sequences with all 8 labels: n=5,066, mean entropy = 0.08 bits, max position = 1.90, 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 = 0.13	n=100,000, mean H = 0.17
`FcRn6`	n=100,000, mean H = 0.13	n=100,000, mean H = 0.16
`R2AH`	n=42,196, mean H = 0.11	n=100,000, mean H = 0.17
`R2AR`	n=88,637, mean H = 0.13	n=100,000, mean H = 0.17
`R2B`	n=99,891, mean H = 0.12	n=100,000, mean H = 0.16
`R3AF`	n=38,090, mean H = 0.12	n=100,000, mean H = 0.17
`R3AV`	n=38,855, mean H = 0.11	n=100,000, mean H = 0.17
`R3B`	n=48,904, mean H = 0.10	n=100,000, mean H = 0.16

Per-source breakdown

`costain`

48,939 rows, mean labels/row = 2.21

Receptor	binders	non-binders	missing	missing %	bind % (of called)
`R1`	19,679	15,148	14,112	28.84%	56.51%
`FcRn6`	0	0	48,939	100.00%	—
`R2AH`	0	0	48,939	100.00%	—
`R2AR`	14,102	10,121	24,716	50.50%	58.22%
`R2B`	15,230	8,806	24,903	50.89%	63.36%
`R3AF`	4,535	20,403	24,001	49.04%	18.19%
`R3AV`	0	0	48,939	100.00%	—
`R3B`	0	0	48,939	100.00%	—

`dms`

316,623 rows, mean labels/row = 2.33

Receptor	binders	non-binders	missing	missing %	bind % (of called)
`R1`	42,760	55,976	217,887	68.82%	43.31%
`FcRn6`	46,235	31,911	238,477	75.32%	59.16%
`R2AH`	29,679	43,034	243,910	77.03%	40.82%
`R2AR`	52,328	52,933	211,362	66.76%	49.71%
`R2B`	68,554	66,150	181,919	57.46%	50.89%
`R3AF`	22,301	30,942	263,380	83.18%	41.89%
`R3AV`	22,357	26,712	267,554	84.50%	45.56%
`R3B`	39,512	107,488	169,623	53.57%	26.88%

`dms+costain`

859 rows, mean labels/row = 4.39

Receptor	binders	non-binders	missing	missing %	bind % (of called)
`R1`	306	300	253	29.45%	50.50%
`FcRn6`	166	109	584	67.99%	60.36%
`R2AH`	91	79	689	80.21%	53.53%
`R2AR`	372	209	278	32.36%	64.03%
`R2B`	446	221	192	22.35%	66.87%
`R3AF`	72	603	184	21.42%	10.67%
`R3AV`	85	59	715	83.24%	59.03%
`R3B`	23	631	205	23.86%	3.52%

`dms+paper_refiltered`

5,291 rows, mean labels/row = 4.86

Receptor	binders	non-binders	missing	missing %	bind % (of called)
`R1`	1,621	1,555	2,115	39.97%	51.04%
`FcRn6`	2,341	616	2,334	44.11%	79.17%
`R2AH`	1,696	1,325	2,270	42.90%	56.14%
`R2AR`	1,902	1,343	2,046	38.67%	58.61%
`R2B`	2,096	1,828	1,367	25.84%	53.41%
`R3AF`	1,510	1,224	2,557	48.33%	55.23%
`R3AV`	1,616	1,235	2,440	46.12%	56.68%
`R3B`	1,759	2,051	1,481	27.99%	46.17%

`dms+paper_refiltered+costain`

1,319 rows, mean labels/row = 6.14

Receptor	binders	non-binders	missing	missing %	bind % (of called)
`R1`	527	583	209	15.85%	47.48%
`FcRn6`	768	247	304	23.05%	75.67%
`R2AH`	303	320	696	52.77%	48.64%
`R2AR`	817	333	169	12.81%	71.04%
`R2B`	847	357	115	8.72%	70.35%
`R3AF`	289	898	132	10.01%	24.35%
`R3AV`	391	254	674	51.10%	60.62%
`R3B`	148	1,022	149	11.30%	12.65%

`paper_refiltered`

1,083,580 rows, mean labels/row = 1.31

Receptor	binders	non-binders	missing	missing %	bind % (of called)
`R1`	94,734	144,732	844,114	77.90%	39.56%
`FcRn6`	73,148	159,546	850,886	78.53%	31.44%
`R2AH`	10,187	97,229	976,164	90.09%	9.48%
`R2AR`	14,127	140,574	928,879	85.72%	9.13%
`R2B`	8,434	96,854	978,292	90.28%	8.01%
`R3AF`	8,446	166,795	908,339	83.83%	4.82%
`R3AV`	13,822	169,949	899,809	83.04%	7.52%
`R3B`	7,414	215,333	860,833	79.44%	3.33%

`paper_refiltered+costain`

18,085 rows, mean labels/row = 3.76

Receptor	binders	non-binders	missing	missing %	bind % (of called)
`R1`	10,564	5,640	1,881	10.40%	65.19%
`FcRn6`	2,800	2,978	12,307	68.05%	48.46%
`R2AH`	240	2,651	15,194	84.01%	8.30%
`R2AR`	4,989	5,341	7,755	42.88%	48.30%
`R2B`	4,284	5,068	8,733	48.29%	45.81%
`R3AF`	937	12,406	4,742	26.22%	7.02%
`R3AV`	584	3,831	13,670	75.59%	13.23%
`R3B`	48	5,710	12,327	68.16%	0.83%

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

Dataset analysis: fc_retraining_v2