Del2Phen 2024.7.15

Del2Phen, a tool to predict phenotypes associated with copy-number variants based on similar CNVs in patient data.

Del2Phen

Introduction

Del2Phen is a tool for predicting phenotypes for a given copy-number variant (CNV), based on known phenotypes from patients with similar CNVs. Patients are first grouped according to minimum similarity thresholds across multiple CNV properties, including genes and haploinsufficient genes affected by the CNV. Phenotypes are then predicted by the prevalence of phenotypes found in individuals within the group.

This project is part of The Chromosome 6 Project, a research initiative that aims to provide information on rare chromosome 6 aberrations in children to parents and healthcare professionals.

Installation

Del2Phen is written in Python 3 and requires Python >= 3.9, except Python 3.10. We recommend using conda to create a Python 3.12 environment, then installing with pip:

conda create -n del2phen_env python==3.12
conda activate del2phen_env
python -m pip install del2phen

Getting Started

At minimum, Del2Phen requires basic patient CNV and phenotype tabular data.

CNV data

CNV input data must be provided as a 5-column tab-separated file. Each row should specify one CNV from one patient. Patients with multiple CNVs should be listed with the same ID on multiple lines, one CNV per line. The file must include the following headings:

• id: Patient ID of the CNV. The same ID can be used multiple times to indicate that one patient has multiple CNVs.
• chromosome: Chromosome on which the CNV is located. These chromosome names must match the names used in the geneset GTF file used (e.g., be sure whether your GTF file uses 'chrX' or 'X' for the chromosome names).
• start: 1-indexed inclusive CNV start position.
• stop: 1-indexed inclusive CNV stop position.
• copy_number: Integer value of the CNV, e.g., a single copy deletion is 1, a single copy duplication is 3.

Phenotype data

Patient phenotype data must be provided as a tab-separated file, where the first column contains patient IDs with the header id. Each additional column must have as a header either a Human Phenotype Ontology (HPO) phenotype ID (e.g., HP:0001643) or any other identifier which must then be defined as a custom phenotype term. Each entry in each column should be coercible to a Boolean or NA (t, T, true, True, f, F, false, False, 0, 1, NA, or an empty value) representing whether or not the patient exhibits the phenotype. Patients are not required to be present in the phenotype file to be present in the CNV file.

Custom phenotypes

Custom phenotypes (i.e., non-HPO identifiers) in the phenotype data must be defined in a separate tab-separated file with two columns with the headers term_id and label. term_id will be used as the unique identifier per phenotype, and label will be used as a display label. See del2phen/resources/custom_phenotypes.tsv as an example.

Prediction

There are several ways to produce phenotype predictions with Del2Phen. The simplest way to predict phenotypes for a single query CNV is the following:

del2phen -g cnvs.tsv -p phenotypes.tsv -cnv chr6:123456-234567:1 --hi-gene-sim 0.75 -x -op ./

The above command will produce a table of phenotypes predicted to be found in patients with a deletion (copy number of 1) on chromosome 6 from bases 123456 to 234567, and will base predictions only on patients that have at least 75% overlap in affected haploinsufficient genes. No other predictions for patients in cnvs.tsv will be produced due to using -x / --cnv-query-only, and results for the queried CNV will be written in cnv_query_predictions.tsv in the directory specified in -op / --output-predictions. Note that without specifying -op / --output-predictions, no predictions will be made.

Phenotypes can be predicted for one hypothetical individual with multiple CNVs by using the -cnv / --cnv-predict flag multiple times:

del2phen -g cnvs.tsv -p phenotypes.tsv -cnv chr6:123456-234567:1 -cnv chrX:1357911-2468101:1 --hi-gene-sim 0.75 -x -op ./

To predict phenotypes for multiple individuals with different CNVs, add additional entries for these individuals to your cnvs.tsv file with a unique identifier for each individual. Do not use the -x / --cnv-query-only flag and do use the --keep-unphenotyped flag to predict phenotypes for these individuals:

del2phen -g cnvs.tsv -p phenotypes.tsv --keep-unphenotyped --hi-gene-sim 0.75 -op ./

If -x / --cnv-query-only is not specified and -op / --output-predictions is, predictions will be made for all patients in -g/--genotypes even if they already have phenotypes described in -p/--phenotypes. By specifying -os / --output-stats with an output file, precision metrics comparing the predicted phenotypes to the known phenotypes in -p/--phenotypes will be written per patient in a single table. This is useful for establishing effective comparison thresholds. You can omit -op / --output-predictions to avoid excessive prediction tables while still outputting prediction metrics for testing:

del2phen -g cnvs.tsv -p phenotypes.tsv --hi-gene-sim 0.75 -os prediction_metrics.tsv

Configuration

Filtering settings

There are several filtering options for subsetting which patients can be used for phenotype prediction to allow flexibility in analysis without the need to directly edit the -g/--genotypes and -p/--phenotypes TSV files.

• A text file with one patient ID per line can be provided with -d / --drop-list to specify patients to completely ignore.
• By default, Del2Phen will consider all CNVs per patient when comparing to the queried patient/CNV(s). For example, querying -cnv chr2:100-200:1 --loci-similarity 0.8 will match a patient in -g/--genotypes with a CNV chr2:100-200:1, as they overlap 100%.
• However, this will not match a patient that has both a CNV chr2:100-200:1 and a CNV chr3:400-500:1, as they share only 50% of their total potentially-shared loci, which falls below the specified threshold of 80%. Comparisons can be restricted to consider only specific chromosomes/contigs by specifying -c / --included-contigs, followed by one or more contig names, space-separated.
• Similarly, analysis can be restricted to only consider specific copy-number CNVs using -cn / --included-copy-numbers, followed by one or more copy number integers, space-separated.

Genotype similarity settings

There are 5 criteria available for comparing patients to one another based on their genotypic similarity, i.e., how similar their CNVs are. If a comparison patient meets all the specified criteria when compared to the queried patient/CNV(s), then the comparison patient can be used to predict phenotypes.

• Length: Require a minimum overlap in the combined total length of CNVs for patients to be compared using --length-sim. This is calculated as the Jaccard index between the total CNV lengths of each pair of patients being compared. Default=0, range=[0, 1].
• Loci: Require a minimum overlap in the combined loci of CNVs for patients to be compared using --loci-sim. This is calculated as the Jaccard index between the total sets of affected loci for each pair of patients being compared. Default=0, range=[0, 1].
• Affected genes: Require a minimum overlap in the genes affected by the CNV of each patient using --gene-sim. This is calculated as the Jaccard index between each patient's set of CNV-affected genes. Default=0, range=[0, 1].
• Affected haploinsufficient genes: Require a minimum overlap in the predicted haploinsufficient genes affected by the CNV of each patient using --hi-gene-sim. This is calculated as the Jaccard index between each patient's set of CNV-affected haploinsufficient genes. Default=0, range=[0, 1].
• Affected dominant-effect genes: Users can specify genes to be considered differently from other genes using -de / --dominant-effect-genes with a space-separated list of gene IDs, or using --de-file to specify a text file containing gene IDs, one per line. Specifying --de-file will override -de / --dominant-effect-genes if it is also present. By default, a comparison patient can only be used to make predictions if they share the same exact set of affected specified genes with the query patient/CNV(s) (note: if neither the query nor the comparison patient have any of these genes affected, they are considered a match and satisfy this criteria). Using the --allow-de-gene-mismatch flag disables this criteria. Alternatively, not specifying any of these genes achieves the same effect. See DE genes and References 1-7 for more information.

Gene settings

HI genes

Del2Phen includes 3 haploinsufficiency (HI) prediction metrics from published studies: HI score, pHaplo score, and pLI score. By default, genes in the provided gene set are considered haploinsufficient if they meet the default criteria of haploinsufficiency from at least one of the 3 metrics, but also requires that at least 2/3 metrics agree if all 3 metrics are available for a gene. The default thresholds of each HI metric are taken from their original studies; however, these thresholds can be altered at runtime by using -pli / --pli-threshold, -hi / --hi-threshold, and -phaplo / --phaplo-threshold. The requirement for how many metrics must agree / how many metrics must be available for a gene to be considered haploinsufficient can be altered using -m / --hi-mode with the following options:

• confirm: Default behavior as defined above.
• any: Only require that any one metric be available any satisfy the threshold for haploinsufficiency.
• 2: Require that at least 2 metrics must be available and that two satisfy their thresholds for haploinsufficiency.
• all: Require that all 3 metrics must be available and that all meet their thresholds for haploinsufficiency.

DE genes

Del2Phen can group patients based on the presence of genes that produce highly penetrant phenotypes when affected by a CNV (called "dominant-effect genes", or DE genes, here). By default, Del2Phen includes an example list of several such DE genes on chromosome 6 (See References 1-7). To provide your own list of these genes, use -de / --dominant-effect-genes to list their gene IDs on the command line, or --de-file to provide a file listing them. To prevent any grouping by DE genes, use --allow-de-gene-mismatch.

Phenotype prediction thresholds

Once a group of patients satisfying the requirements of the genotype similarity settings is identified, phenotypes from this group are only predicted if they also meet minimum prevalence thresholds, which can be set with the following options:

• An absolute minimum number of patients in the group that exhibit the phenotype can be specified with -abs / --absolute-threshold (default=2).
• A relative minimum proportion of patients in the group that exhibit the phenotype can be specified with -rel / --relative-threshold (default=0.2).
  • This proportion is calculated, by default, as the number of patients that exhibit the phenotype (True response) out of all identified patients (True, False, and NA responses). This behavior can be modified with --ignore-nas to ignore patients with an NA response, such that the proportion is calculated as True out of True + False responses only, within the identified patient group.
• A minimum group size can be required, such that if an insufficient number of genotypically-similar patients are discovered, no phenotypes are reported, using --group-size.

Reference gene information

Del2Phen requires reference genome information to compare the gene content of CNVs between patients. This information includes the location of each gene and haploinsufficiency prediction scores.Del2Phen is packaged with 4 reference files in del2phen/resources/ and uses these by default:

• hg19.ensGene.transcripts.gtf.gz: Gene locus information from Ensembl for human genome build 19 (hg19)
• gnomad.v2.1.1.lof_metrics.by_gene.tsv: pLI haploinsufficiency scores for hg19
• HI_Predictions.v3.bed: HI haploinsufficiency scores for hg19
• phaplo.tsv: pHaplo haploinsufficiency scores for hg19

Other reference files can be provided using --gtf-file, --pli-file, --hi-file, and --phaplo-file. Ensure that the gene IDs and genome builds are consistent.

NOTE: To prevent inadvertent mixing of genome builds/reference files, using any of the above arguments will stop Del2Phen from using any of the pre-packaged default files. If, for example, you supply your own updated pHaplo score file, you must also provide, at minimum, a GTF file; analysis will proceed without the other two haploinsufficiency score files, but no pLI or HI metrics will then be included in the analysis if they are not provided.

References

1. The phenotypic spectrum of terminal and subterminal 6p deletions based on a social media-derived cohort and literature review - Rraku, et al., Orphanet Journal of Rare Diseases 2023 - Link
2. The phenotypic spectrum of terminal 6q deletions based on a large cohort derived from social media and literature: a prominent role for DLL1 - Engwerda et al., Orphanet Journal of Rare Diseases 2023 - Link
3. TAB2 deletions and variants cause a highly recognisable syndrome with mitral valve disease, cardiomyopathy, short stature and hypermobility - Engwerda et al., European Journal of Human Genetics 2021 - Link
4. SYNGAP1-Related Intellectual Disability - Holder et al., GeneReviews 2019 - Link
5. Coffin-Siris Syndrome - Vergano et al., GeneReviews 2013 - Link
6. The ARID1B phenotype: what we have learned so far - Santen et al., American Journal of Medical Genetics 2014 - Link
7. Haploinsufficiency of the Notch Ligand DLL1 Causes Variable Neurodevelopmental Disorders - Fischer-Zirnsak et al., American Journal of Human Genetics 2019 - Link
8. Characterising and Predicting Haploinsufficiency in the Human Genome - Huang et al., PLoS Genetics, 2010 - Link
9. Analysis of protein-coding genetic variation in 60,706 humans - Lek et al., Nature, 2016 - Link
10. A cross-disorder dosage sensitivity map of the human genome - Collins et al., Cell, 2022 - Link