Our paper “Data-driven design of targeted gene panels for estimating immunotherapy biomarkers” puts forward a framework for choosing which genes to include in an immunotherapy-focused gene panel, and for estimating from those genes’ mutation data metrics that can inform clinical decisions. One of our method’s key advantages is its flexibility: our framework permits practitioners to choose a target biomarker of interest, fix a maximum panel size, and to incorporate restrictions on whether to include or exclude particular genes.

The biomarkers we consider are exome-wide mutation counts. This class includes the standard companion prognostic biomarker for immunotherapy, Tumour Mutation Burden (TMB). In our paper we demonstrate our method’s performance predicting TMB and classifying TMB high/low tumours. We also show that our method is effective in predicting other biomarkers such as Tumour Indel Burden (TIB).

In this blog post we’ll give an overview of how these biomarkers are used to enable precision oncology, the challenges faced by practitioners, and the main contributions of our work. We invite you to read our paper and make use of our open-source R package.

Why use biomarkers?

The goal of cancer genomics research is to use sequencing technology to make informed decisions about patient care. In our case, the decision is whether to prescribe Immune Checkpoint Blockade (ICB) therapy to a cancer patient, based on the mutations observed in their tumour. ICB is a treatment that restricts natural inhibitors of the immune system to allow a fuller anti-tumour response. Many patients experience durable benefit from ICB therapy, but it doesn’t work for everyone. 

By defining an exome-wide biomarker such as TMB or TIB, we use knowledge from cancer immunology to inform how we use patients’ mutation data to predict the effectiveness of ICB. Specifically, we leverage the known relationship between coding mutations, production of cell-surface neoantigens, and immune surveillance. The aim is that a good biomarker should reflect how easily the immune system can recognise tumour cells as ‘foreign’. 

While immunotherapy biomarkers do not contain all the possible relevant information to predict whether a patient will respond to treatment, they are a key quantity to be considered by clinicians, and amongst the most promising steps towards precision oncology. Indeed, TMB can be used as a standalone predictor of response to immunotherapy, or collectively alongside other clinical variables such as PDL-1 expression and tumour lymphocyte infiltration. In either case, the simple interpretation derived from TMB measurements allows for more informed decision making.

Why use targeted gene panels?

Having elected to use a biomarker like TMB to extract clinically relevant information, we need to consider how this can be implemented in the real world. A major issue facing practitioners is the cost of tumour whole-exome sequencing, which is still prohibitive for the majority of patients. To address this, multiple groups have designed concise gene panels from which to estimate TMB. These approaches have comprised a blend of manual curation and pre-defined statistical selection criteria.  

The challenge here is to balance the improvement in predictive performance from adding a gene to a given panel against the additional associated cost. To address this, we start with an assumption that the cost of deploying a targeted gene panel is proportional to the total coding length of the genomic regions sequenced. This approximation provides a useful baseline, and can be adapted easily by the user to specify their own cost metric. Furthermore, in practice gene panels are typically designed with multiple goals in mind, and so our methodology is designed to facilitate adding genes to an existing panel to improve predictive performance for a particular biomarker.