Table 1 Summary of trained modules and models
From: MMSplice: modular modeling improves the predictions of genetic variant effects on splicing
MMSplice model | Training data | Architecture | Loss function | Target value | Parameters |
|---|---|---|---|---|---|
Donor module | GENCODE 24, positive: annotated donors, negative: random sequence (“Methods” section) | Four layer neural network with dropout and batch normalization, Additional file 1: Figure S1A | Binary cross entropy | Positive vs. negative | 18,049 |
Acceptor module | GENCODE 24, positive: annotated acceptors, negative: random sequence (“Methods” section) | Two layer conv. neural network with dropout and batch normalization, Additional file 1: Figure S1B | Binary cross entropy | Positive vs. negative | 4833 |
Exon 5 ′ module | MPRA [18] exonic sequence | One conv. layer shared with the Exon 3 ′ module, followed with one specific dense layer, Additional file 1: Figure S2 | Binary cross entropy | Ψ 5 | 6145 |
Exon 3 ′ module | MPRA [18] exonic sequence | One conv. layer shared with the Exon 5 ′ module, followed with one specific dense layer, Additional file 1: Figure S2 | Binary cross entropy | Ψ 3 | 6145 |
Intron 5 ′ module | MPRA [18] intronic sequence | One conv. layer shared with the Intron 3 ′ module, followed with one specific dense layer, Additional file 1: Figure S2 | Binary cross entropy | Ψ 3 | 13,825 |
Intron 3 ′ module | MPRA [18] intronic sequence | One conv. layer shared with the Intron 5 ′ module, followed with one specific dense layer, Additional file 1: Figure S2 | Binary cross entropy | Ψ 5 | 13,825 |
Δlogit(Ψ) model | Vex-seq [29] | Linear regression | Huber loss | Δlogit(Ψ), Eq. 2 | 9 |
Splicing efficiency model (in vivo) | MaPSy (“Methods” section) | Linear regression | Huber loss | Splicing efficiency, Eq. 10 | 5 |
Splicing efficiency model (in vitro) | MaPSy (“Methods” section) | Linear regression | Huber loss | Splicing efficiency, Eq. 10 | 5 |
Pathogenicity model (w/o phyloP and CADD) | ClinVar [30] [ − 10, 10] around donor, [ − 40, 10] around acceptor | Logistic regression | Binary cross entropy | Pathogenic vs. benign | 14 |
Pathogenicity model (with phyloP and CADD) | ClinVar [30] [ − 10, 10] around donor, [ − 40, 10] around acceptor | Logistic regression | Binary cross entropy | Pathogenic vs. benign | 18 |