There is a clear unmet need in availability of a clinically applicable classifier for the prospective classification of tumour samples for immune checkpoint therapy, especially when applied to formalin-fixed, paraffin embedded (FFPE) material. To this end, we developed and analytically validated a novel cross-platform web-based classification model using a comprehensive machine learning approach. This classification model could be used prospectively in the clinic to stratify patients into six identified immune subtypes. In contrast to the traditional immune subgrouping provided by the PanCan Atlas consortium, which assigns all tumours to one of six immune subtypes, we provide an objective measure by which ‘predominant subgroups’ and ‘non-classifiable’ cases can be also identified. The developed model has achieved a performance of 96.2% balanced accuracy on more than 2000 solid tumours from different malignancies.
|Media of output||Bioinformatics Tool|
|Publication status||Early online date - 2019|