model-template.yaml

type: keras
args:
  arch:
    url: {{ args_arch_url }}
    md5: {{ args_arch_md5 }}
  weights:
    url: {{ args_weights_url }}
    md5: {{ args_weights_md5 }}
  backend: tensorflow
  image_dim_ordering: tf
info:
  authors:
      - name: Haoyang Zeng
        github: haoyangz
  contributors:
      - name: Roman Kreuzhuber
        github: krrome
  name: CpGenie
  version: 0.1
  trained_on: "Samples from Chromosome 1-9 and 14-22 were used for training, 12-13 were used for hyper-parameter tuning and model selection, and the rest of the data were held-out for testing. In the analysis of variant-prediction, only meQTL and allele-specific methylation data from the held-out chromosome 10 and 11 were used. The 50 RRBS datasets of immortal cell lines, including GM12878, and the WGBS dataset of GM12878 were downloaded from ENCODE website (https://www.encodeproject.org/)."
  doc: >
    Abstract:
    DNA methylation plays a crucial role in the establishment of tissue-specific
    gene expression and the regulation of key biological processes. However, our
    present inability to predict the effect of genome sequence variation on DNA
    methylation precludes a comprehensive assessment of the consequences of
    non-coding variation. We introduce CpGenie, a sequence-based framework that
    learns a regulatory code of DNA methylation using a deep convolutional neural
    network and uses this network to predict the impact of sequence variation on
    proximal CpG site DNA methylation. CpGenie produces allele-specific DNA methylation
    prediction with single-nucleotide sensitivity that enables accurate prediction
    of methylation quantitative trait loci (meQTL). We demonstrate that CpGenie
    prioritizes validated GWAS SNPs, and contributes to the prediction of
    functional non-coding variants, including expression quantitative trait
    loci (eQTL) and disease-associated mutations. CpGenie is publicly available to
    assist in identifying and interpreting regulatory non-coding variants.
  cite_as: https://doi.org/10.1093/nar/gkx177
  license: Apache License v2
  trained_on: RRBS (restricted representation bisulfite sequencing) data from ENCODE (https://www.encodeproject.org/)
  training_procedure: RMSprop
  tags:
    - DNA methylation
default_dataloader:
  defined_as: kipoiseq.dataloaders.SeqIntervalDl
  default_args:
    auto_resize_len: 1001
    alphabet_axis: 0
    dummy_axis: 1
dependencies:
  conda:
    - python=3.6.12
    - h5py=2.10.0
    - tensorflow=1.10.0
    - keras=1.2.2
    - pysam=0.15.3
    - pip=20.2.4
schema:
  inputs:
    name: seq
    special_type: DNASeq
    shape: (4, 1, 1001)
    doc: DNA sequence
    associated_metadata: ranges
  targets:
    name: methylation_prob
    shape: (2, )
    doc: Methylated and Unmethylated probabilities
    column_labels:
            - methylation_prob
            - unmethylation_prob
postprocessing:
    variant_effects:
      seq_input:
        - seq
      use_rc: True
      scoring_functions:
        - type: diff
          default: true
        - type: logit
          default: true
{% if model == 'GM19239_ENCSR000DGH' %}
test:
  expect:
    url: https://s3.eu-central-1.amazonaws.com/kipoi-models/predictions/14f9bf4b49e21c7b31e8f6d6b9fc69ed88e25f43/CpGenie/{{ model }}/predictions.h5
    md5: 3727f6763f5c8cc2f8a40889c84cf123
  precision_decimal: 6
{% endif %}