Using ALoFT to determine the impact of putative loss-of-function variants in protein-coding genes

Suganthi Balasubramanian, Yao Fu, Mayur Pawashe, Patrick D. McGillivray, Mike Jin, Jeremy Liu, Konrad J. Karczewski, Daniel G. MacArthur, Mark GersteinNature Communications 2017 · DOI

Abstract

Variants predicted to result in the loss of function of human genes have attracted interest because of their clinical impact and surprising prevalence in healthy individuals. Here, we present ALoFT (annotation of loss-of-function transcripts), a method to annotate and predict the disease-causing potential of loss-of-function variants. Using data from Mendelian disease-gene discovery projects, we show that ALoFT can distinguish between loss-of-function variants that are deleterious as heterozygotes and those causing disease only in the homozygous state. Investigation of variants discovered in healthy populations suggests that each individual carries at least two heterozygous premature stop alleles that could potentially lead to disease if present as homozygotes. When applied to de novo putative loss-of-function variants in autism-affected families, ALoFT distinguishes between deleterious variants in patients and benign variants in unaffected siblings. Finally, analysis of somatic variants in >6500 cancer exomes shows that putative loss-of-function variants predicted to be deleterious by ALoFT are enriched in known driver genes. Variants causing loss of function (LoF) of human genes have clinical implications. Here, the authors present a method to predict disease-causing potential of LoF variants, ALoFT (annotation of Loss-of-Function Transcripts) and show its application to interpreting LoF variants in different contexts.

Introduction

Every human genome contains over 100 putative loss-of-function (pLoF) variants—mutations like premature stop codons or frameshifts that appear to "break" a gene. Surprisingly, many of these occur in healthy individuals. The challenge for geneticists is distinguishing the "benign" pLoFs, which the body might tolerate through redundant pathways or alternative splicing, from "deleterious" ones that cause disease.

Existing tools often struggle with pLoFs because they focus on missense mutations (single amino acid changes) rather than total protein truncation. Furthermore, current methods rarely account for zygosity: whether a mutation is harmful when only one copy is broken (dominant) or only when both copies are affected (recessive). ALoFT (Annotation of Loss-of-Function Transcripts) was built to fill this gap, providing a machine-learning pipeline designed to predict whether gene-truncating mutations are benign or disease-causing by distinguishing between dominant and recessive modes of inheritance.

Methods

ALoFT is a comprehensive annotation and prediction pipeline using a Random Forest classifier trained on three distinct classes of variants: benign common homozygous pLoF variants, recessive disease-causing mutations, and dominant disease-causing mutations in haploinsufficient genes.

The ALoFT pipeline: from VCF input through multi-feature annotation to pathogenicity prediction and final output.

The pipeline integrates 108 different features across several categories:

  • Functional: Removal of known PFAM/SMART domains or signal peptides.
  • Evolutionary: Conservation scores (GERP) and dN/dS ratios.
  • Network-based: The gene’s proximity to known disease genes in protein-protein interaction networks.
  • Transcript-specific: Predictions of Nonsense-Mediated Decay (NMD)—the cell's quality control mechanism.

Datasets

The study utilized a massive array of genomic data for training and validation, including the 1000 Genomes Project (1KG), HGMD, and ClinVar. Population frequency data was sourced from ExAC and ESP6500. For disease applications, the authors analyzed de novo mutations from autism-affected families (Simons Simplex Collection), somatic mutations from over 6,500 cancer exomes (TCGA), and data from the Center for Mendelian Genomics (CMG).

Key Insights and Results

ALoFT achieved high accuracy with an average multiclass AUC of 0.97. It demonstrated high precision across all categories: Benign (0.96), Recessive (0.86), and Dominant (0.86).

Network and Contextual Nuance

Genes that act as "hubs" or are closely connected to other disease genes in biological networks are significantly more likely to harbor dominant, deleterious pLoFs. Additionally, ALoFT successfully identifies rare, pathogenic "last-exon" mutations. While these are often assumed to be benign because they escape NMD, ALoFT identifies those that produce toxic truncated proteins.

Clinical Applications

  • Autism: Patients had a significantly higher proportion of predicted "dominant" deleterious pLoFs compared to their unaffected siblings ($p = 8.4 \times 10^{-4}$).
  • Cancer: pLoFs predicted as deleterious were heavily enriched in known driver genes; 87% of pLoFs in tumor suppressor genes (meeting the "20/20 rule") were classified as deleterious.
  • Genetic Landmines: The study estimates that every healthy individual carries at least two rare heterozygous premature stop alleles that would likely cause disease if present in a homozygous state.

Comparison with Existing Tools

ALoFT significantly outperformed general pathogenicity scores like CADD and GERP in distinguishing between dominant and recessive disease variants. While tools like CADD correctly identify these variants as deleterious, they lack the multiclass specificity to distinguish the mode of inheritance.

Comparison of CADD scores for dominant and recessive disease variants; while both show high pathogenicity scores, CADD does not distinguish between the two modes of inheritance.

Limitations and Future Work

ALoFT is specifically designed for loss-of-function and does not currently model "gain-of-function" mutations, where a truncated protein acquires new toxic activity. While it works on frameshift indels, the training was primarily based on nonsense mutations due to the reliability of older datasets. Future work will involve correlating ALoFT scores with high-throughput experimental assays to further refine the model's predictive power.

Summary updated

Abstract

Variants predicted to result in the loss of function of human genes have attracted interest because of their clinical impact and surprising prevalence in healthy individuals. Here, we present ALoFT (annotation of loss-of-function transcripts), a method to annotate and predict the disease-causing potential of loss-of-function variants. Using data from Mendelian disease-gene discovery projects, we show that ALoFT can distinguish between loss-of-function variants that are deleterious as heterozygotes and those causing disease only in the homozygous state. Investigation of variants discovered in healthy populations suggests that each individual carries at least two heterozygous premature stop alleles that could potentially lead to disease if present as homozygotes. When applied to de novo putative loss-of-function variants in autism-affected families, ALoFT distinguishes between deleterious variants in patients and benign variants in unaffected siblings. Finally, analysis of somatic variants in >6500 cancer exomes shows that putative loss-of-function variants predicted to be deleterious by ALoFT are enriched in known driver genes.Variants causing loss of function (LoF) of human genes have clinical implications. Here, the authors present a method to predict disease-causing potential of LoF variants, ALoFT (annotation of Loss-of-Function Transcripts) and show its application to interpreting LoF variants in different contexts.
BibTeX
@inproceedings{balasubramanian2017using,
  author = {Suganthi Balasubramanian and Yao Fu and Mayur Pawashe and Patrick D. McGillivray and Mike Jin and Jeremy Liu and Konrad J. Karczewski and Daniel G. MacArthur and Mark Gerstein},
  title = {{Using ALoFT to determine the impact of putative loss-of-function variants in protein-coding genes}},
  booktitle = {Nature Communications},
  year = {2017},
  doi = {10.1038/s41467-017-00443-5},
}
Figure