Take Note!

Our services require cookies to keep track of your submissions and find your results. Submissions expire after 24 hours.

What is BadMut?

BadMut (Bad Mutation; we like how it resembles "bad mood") is a meta-estimator designed to distinguish between neutral and disease-causing amino acid substitutions (i.e. nonsynonimous single nucleotide variations – nsSNVs). It is based on a machine-learning technique commonly known as deep learning, that is it uses deep (multilayer) artificial neural networks to make predictions.

What is a meta-estimator?

Untangling the connections between variation and phenotypic traits remains the greatest challenge of functional genomics, thus a lot of effort has been put into developing the means to infer possible damage of novel (phenotypically uncategorised) nsSNVs by employing machine-learning. As a result, over the past decade many algorithms have been developed for predicting deleteriousness. In order to make predictions these tools encode variants using multiple quantitative and qualitative features of proteins and their sequences. An ensemble of such tools can be further used to get an integrated prediction - this is exactly what meta-estimators (aka ensemble scores and meta-scores) do.

What kind of information BadMut uses to make predictions?

Being a meta-estimator, BadMut uses the output of other scores as its main source of information. In particular, BadMut integrates SIFT, Polyphen-2, LRT, MutationTaster, MutationAssessor, FATHMM, PROVEAN, MetaSVM, MetaLR, GERP++, phyloP, phastCons, SiPhy. It also uses population statistics (allele frequency) generated by the 1000 Genomes project. BadMut takes all the information from dbNSFP v3.1.

What can one use BadMut for?

BadMut can be used to prioritise likely-damaging novel amino acid substitutions in silico prior to expensive laboratory testing. Although it doesn't analyse noncoding regions of the genome, unlike most other tools it provides near-complete coverage of the human exome.

How to interpret the output?

Unlike many other tools BadMut doesn't assign the substitutions it processes to different classes (e.g. pathogenic/neutral). Instead, it returns the probability of clinical significance. One thus can apply a binary cutoff threshold to separate neutral and pathogenic substitutions. Selecting a value for the threshold is a question of balancing sensitivity (the ability to find damaging mutations) and specificity – lower thresholds result in greater sensitivity and lower specificity and vice versa.

This tab allows you to submit a single allele at a time. While you can make as many of these submissions as you want, if you have more than a dozen of alleles to process, you should consider the VCF submission form instead.

You can upload and process any VCF file (v4.0-4.2), though your upload can be no more than 20MB. The file can be compressed by any gzip-compatible method (for instance, gzip itself or bgzip). To limit workload, our system only processes the first 500k lines from a submitted VCF file (excluding the header lines). The output is a bgzipped VCF v4 file.