DeepTrio training data
February 11, 2026 ยท View on GitHub
WGS models
| version | Replicates | #examples |
|---|---|---|
| Child model | ||
| 1.1.0 | 4 HG001/NA12891/NA12892 trios 7 HG005/HG006/HG007 trios 3 HG002/HG003/HG004 trios | 566,589,652(1) |
| 1.2.0 | (Same model as 1.1.0) | |
| 1.3.0 | (Same model as 1.1.0) | |
| 1.4.0 | 4 HG001/NA12891/NA12892 trios 7 HG005/HG006/HG007 trios 3 HG002/HG003/HG004 trios | 704,228,446 |
| 1.5.0 | (6)4 HG001, 3 HG002, 3 HG003, 3 HG004, 7 HG005, 6 HG006, 6 HG007, 4 NA12891, 4 NA12892 | 704,228,358 |
| 1.6.0 | (6)4 HG001, 3 HG002, 3 HG003, 3 HG004, 7 HG005, 6 HG006, 6 HG007, 4 NA12891, 4 NA12892 | 704,394,556 |
| Parent model | ||
| 1.1.0 | 7 HG005/HG006/HG007 trios 3 HG002/HG003/HG004 trios | 315,847,934 |
| 1.2.0 | (Same model as 1.1.0) | |
| 1.3.0 | (Same model as 1.1.0) | |
| 1.4.0 | 7 HG005/HG006/HG007 trios 3 HG002/HG003/HG004 trios | 457,374,516 |
| 1.5.0 | (6)3 HG002, 3 HG003, 3 HG004, 7 HG005, 6 HG006, 6 HG007 | 457,374,464 |
| 1.6.0 | (6)2 HG001, 3 HG002, 3 HG003, 3 HG004, 7 HG005, 6 HG006, 6 HG007, 2 NA12891, 2 NA12892 | 457,420,038 |
| 1.9.0 | (6)2 HG001, 3 HG002, 3 HG003, 3 HG004, 7 HG005, 6 HG006, 6 HG007, 2 NA12891, 2 NA12892 | 457,420,038 |
| 1.10.0 | (6)2 HG001, 3 HG002, 3 HG003, 3 HG004, 7 HG005, 6 HG006, 6 HG007, 2 NA12891, 2 NA12892 | 457,420,038 |
WES models
| version | Replicates | #examples |
|---|---|---|
| Child model | ||
| 1.1.0 | 27 HG001/NA12891/NA12892 trios 6 HG005/HG006/HG007 trios 7 HG002/HG003/HG004 trios | 18,002,596 |
| 1.2.0 | (Same model as 1.1.0) | |
| 1.3.0 | (Same model as 1.1.0) | |
| 1.4.0 | 27 HG001/NA12891/NA12892 trios 6 HG005/HG006/HG007 trios 6 HG002/HG003/HG004 trios | 27,776,416 |
| 1.5.0 | (6)9 HG001, 7 HG002, 7 HG003, 7 HG004, 8 HG005, 8 HG006, 8 HG007, 9 NA12891, 9 NA12892 | 27,791,954 |
| 1.6.0 | (6)9 HG001, 7 HG002, 7 HG003, 7 HG004, 8 HG005, 8 HG006, 8 HG007, 9 NA12891, 9 NA12892 | 27,783,324 |
| Parent model | ||
| 1.1.0 | 6 HG005/HG006/HG007 trios 6 HG002/HG003/HG004 trios | 4,131,018 |
| 1.2.0 | (Same model as 1.1.0) | |
| 1.3.0 | (Same model as 1.1.0) | |
| 1.4.0 | 6 HG005/HG006/HG007 trios 6 HG002/HG003/HG004 trios | 13,036,995 |
| 1.5.0 | (6)6 HG002, 6 HG003, 6 HG004, 8 HG005, 8 HG006, 8 HG007 | 13,036,998 |
| 1.6.0 | (6)6 HG002, 6 HG003, 6 HG004, 8 HG005, 8 HG006, 8 HG007 | 13,039,595 |
| 1.9.0 | (6)6 HG002, 6 HG003, 6 HG004, 8 HG005, 8 HG006, 8 HG007 | 13,039,595 |
| 1.10.0 | (6)6 HG002, 6 HG003, 6 HG004, 8 HG005, 8 HG006, 8 HG007 | 13,039,595 |
PACBIO models(2)(3)
| version | Replicates | #examples |
|---|---|---|
| Child model | ||
| 1.1.0 | 1 HG005/HG006/HG007 trio 8 HG002/HG003/HG004 trios | 397,610,700 |
| 1.2.0 | 1 HG005/HG006/HG007 trio 8 HG002/HG003/HG004 trios | 406,893,180(4) |
| 1.3.0 | 2 HG005/HG006/HG007 trio 10 HG002/HG003/HG004 trios | 539,382,124(5) |
| 1.4.0 | (Same model as 1.3.0) | |
| 1.6.0 | 9 HG002, 5 HG003, 5 HG004, 1 HG005, 1 HG006, 1 HG007 | 890,016,014(5) |
| Parent model | ||
| 1.1.0 | 1 HG005/HG006/HG007 trio 8 HG002/HG003/HG004 trios | 386,418,918 |
| 1.2.0 | 1 HG005/HG006/HG007 trio 8 HG002/HG003/HG004 trios | 392,749,204(4) |
| 1.3.0 | 2 HG005/HG006/HG007 trio 10 HG002/HG003/HG004 trios | 533,353,050(5) |
| 1.4.0 | (Same model as 1.3.0) | |
| 1.6.0 | 9 HG002, 5 HG003, 5 HG004, 1 HG005, 1 HG006, 1 HG007 | 838,515,085(5) |
| 1.9.0 | 9 HG002, 5 HG003, 5 HG004, 1 HG005, 1 HG006, 1 HG007 | 607,118,560(5) |
| 1.10.0 | 9 HG002, 5 HG003, 5 HG004, 1 HG005, 1 HG006, 1 HG007 | 607,118,560(5) |
ONT models(2)(3)
| version | Replicates | #examples |
|---|---|---|
| 1.6.0 | 1 HG002, 1 HG002, 1 HG004 | 50,249,704(5) |
| Parent model | ||
| 1.6.0 | 1 HG002, 1 HG002, 1 HG004 | 99,675,190(5) |
| 1.9.0 | 5 HG002, 5 HG004, 4 HG005, 4 HG006, 4 HG007 | 607,118,560(5) |
| 1.10.0 | 5 HG002, 5 HG004, 4 HG005, 4 HG006, 4 HG007 | 607,118,560(5) |
(1): We include HG002/HG003/HG004 for training WGS model, but only using examples from the region of NIST truth confident region v4.2 subtracting v3.3.2.
(2): We use the entire HG002/HG003/HG004 trio for PacBio model training.
(3): PacBio and ONT training data contains training examples with haplotag sorted images.
(4): In v1.2.0, we updated the NIST truth versions we used for training.
(5): In v1.3.0, we included PacBio Sequel II Chemistry v2.2 data in the training dataset. And we updated to NIST truth version to v4.2.1.
(6): Starting in v1.5.0, for clarity, we report the number of unique BAM files used. Note that this doesn't mean all the trios were paired together to produce training data.