Using AI variant interpretation tools

In recent years, state-of-the-art artificial intelligence (AI) has been applied to improve the variant interpretation process. Implementing AI algorithms is especially helpful when a large amount of data needs to be processed to arrive at an answer – therefore making it very applicable to variant analysis. Using AI-enabled tools for variant interpretation might also be helpful when working in a resource-limited setting, or on interpreting variants from diverse populations where the data resources might be sparse and sporadic.

AI can be subdivided into different fields, including natural language processing, machine learning and deep learning. Different tools may use the same or different AI processes. It is important to consider that there isn’t an optimal approach for everything – one should consider the mechanisms of the tools and the purpose of the research to choose an approach accordingly.

DeepVariant (developed by Google) and DNAscope are two examples of AI-based variant callers that may be used instead of more conventional tools, such as BCFTools. A recent evaluation of these tools showed that the AI-based tools outperformed conventional tools for calling SNVs and INDELs. These results held true for both short-read and long-read generation sequencing data. AI-driven variant prioritisation pipelines have also been shown to be a highly efficient approach that can accelerate the discovery of clinically relevant genetic variants. Franklin by Genoox is an example of an extensive AI-based evidence network of genomic and clinical data, which is available freely to the scientific community.

AI-based tools for detection and prioritisation of disease-causative variants

Variant calling

Single Nucleotide Variants (SNPs)/indels

HMM: Hidden Markov Model; CNN: Convolutional Neural Networks

Copy Number Variants (CNVs)

Variant prioritising

Coding variants

SVM: Support Vector Machine

Splicing variants

Regulatory variants

GLM-LARS: Generalised Linear Model-based Least Angle Regression

Currently, there are AI tools to perform each step of the variant interpretation workflow (as illustrated in the Table above). Choosing which tool to use may depend on your setting and resources. These tools may be especially useful when you have limited bioinformatics support, as they are user-friendly and intuitive. Some tools are proprietary software, which means that there is a cost involved in using these tools, another important consideration, especially in resource-constrained environments. It is important to consider that AI tools are only as good as the data available to train these tools – AI is therefore also limited by the lack of diversity in data available for variant interpretation.

Use the comments to share other AI tools you may know and your experience with this technology for genomics.