Predict

The predict function will use a maxATAC model to predict TF binding in a new condition. The user must provide a model and a bigwig file that corresponds to an ATAC-seq signal track.

Example

maxatac predict --model CTCF.h5 --signal GM12878.bigwig

or

maxatac predict --tf CTCF --signal GM12878.bigwig

Required Arguments

-tf, --tf_name or -m, --model

The user must provide either the TF name that they want to make predictions for OR the h5 model file they desire. If the user provides a TF name, the best model will be used and the correct threshold file will be provided for peak calling.

-s, --signal, -i

The ATAC-seq signal bigWig track that will be used to make TF binding predictions.

-n, --name, --prefix

Output filename prefix (without extension) to use. Default maxatac_predict.

Optional Arguments

--batch_size

The number of regions to predict on per batch. Default 10000. Decrease this value if you are having memory issues.

--bed, --peaks, --regions, , --roi, -roi

The path to a BED file containing genomic regions to focus TF predictions on. These peaks will be used to refine the prediction windows. Default: whole-chromosome predictions.

-bl, --blacklist

The path to a bigWig file that has regions to exclude. Default: maxATAC-defined blacklist.

-c, -chroms, --chromosomes

The chromosomes to make predictions on. Our models do not currently consider chromosomes X or Y. This means that most of the files will not contain this information. You should not predict in chrX or chrY unless you know your bigWig contains these chromosomes. Default: human autosomal chromosomes 1-22. Note: this argument MUST be specified in conjunction with --genome, -sequence, and --chrom_sizes if the input file was aligned to a genome build other than hg38.

-cs, --chrom_sizes, -chrom_sizes, --chromosome_sizes

The path to the chromosome sizes file. This is used to generate the bigwig signal tracks. Note: this argument MUST be specified in conjunction with --genome, -sequence, and --chromosomes if the input file was aligned to a genome build other than hg38.

"-cf", -cutoff_file, --cutoff_file

The cutoff file provided in /data/models that corresponds to the average validation performance metrics for the TF model.

"-ct", "-cutoff_type", "--cutoff_type"

The cutoff type (i.e. Precision, Recall, F1, log2FC). (F1 = F1-score, and log2FC = Log2( Precision : Random Precision)). Default: F1.

"-cv", "-cutoff_value", "--cutoff_value"

The cutoff value for the cutoff type provided. Note: precision, recall, and F1-scores range from 0-1, while better-than-random log2FC scores range from 0 to infinity. Example: 0.7.

--genome

The genome build that was used for alignment of the ATAC-seq signal file. Default: hg38.

--max_zooms

The number of zoom levels that should be computed for the output bigWig file. Zoom levels are pre-computed summary statistics that enable fast zooming into/out of a genomic region in a bigWig file when a visualization tool (e.g., IGV, UCSC Genome Browser). Lower values of this parameter result in slower loading of bigWig files in visualization tools, while higher values of this parameter result in a large memory overhead. The range of potential parameter values is (0-10). Default: 5. Note: if this argument is set to 0, the resulting bigWig files are NOT compatible with other bigWig tools (e.g., deepTools) and cannot be visualized using tools like IGV and the UCSC Genome Browser. Please see: https://github.com/deeptools/pyBigWig/blob/master/README.md for additional details.

--loglevel

This argument is used to set the logging level. Currently, the only working logging level is ERROR.

-o, --output

Output directory path. Default: ./prediction_results

--sequence, --seq

This argument specifies the path to the 2bit DNA sequence for the genome of interest. maxATAC models are trained with hg38, so you will need the correct .2bit file. Note: this argument MUST be specified (with a valid 2bit file) in conjunction with --genome, --chrom_sizes, and --chromosomes if the input file was aligned to a genome build other than hg38.

-skip_call_peaks, --skip_call_peaks

This will skip calling peaks on prediction tracks. Default: False

--step_size

The step size to use for building the prediction intervals. Overlapping prediction bins will be averaged together. Default: INPUT_LENGTH/4, where INPUT_LENGTH is the maxATAC model input size of 1,024 bp.

-w, --windows

The windows to use for prediction. These windows must be 1,024 bp wide and have a consistent step size.