P03.03 Novel machine-learning tools improve cost-effective development of personalised immunotherapies: lowering false positive rates in the search for actionable immunogenic neoantigens

Background Personalised immunotherapy approaches rely on the ability of tumour-derived neoantigens to elicit a T-cell immune reaction able recognise and kill tumour cells expressing them. Clinical attempts leverage power have however yielded mixed results. This can mostly be attributed difficulty finding truly immunogenic peptides from set novel generated by mutations in given cancer patient. In silico help alleviate this heavy cost reducing neoantigen search space, prioritising epitopes based various parameters such as epitope expression or MHC binding likelihood. Here we present suite tools aimed at further assisting clinicians selecting most actionable potential candidates. Materials Methods We developed neoMS, neural network algorithm predict presentation cell surface with unparalleled performance. Furthermore, neoIM is discriminate, an HLA-agnostic fashion, which presented will reaction. first-in-class random forest classifier specifically trained classify short length 9-11 amino acids non-immunogenic. Results The neoMS model achieves up 0.61 precision recall 0.4 its test set, vastly outperforming current industry standards. addition, due his sequence-based comparison method, exhibits extrapolation capabilities, achieving non-zero predictive when evaluated ground truth ligandome data derived HLA allele completely absent training set. some indications showed, moreover, that neoMS-predicted rate used combination mutational burden high-specificity predictor response checkpoint inhibitor treatment. outperforms currently available methods peptide immunogenicity high accuracy (AUC=0.88). Interestingly, confirmed ELISPOT obtained Dillon et al. (2017) showing 4 out 11 breast patients vaccine consisting 9 class-I restricted cancer-associated peptides. 2 antigens resulted CD8+ specific were predicted highest scoring neoantigens. Conclusions Taken together, these decrease false positive rates significantly they enable improved identification predictions correlate intensity clinical benefits. As such, represent cost-efficient preliminary step for actionable, Reference PM, Petroni GR, Smolkin ME, Brenin DR, Chianese-Bullock KA, Smith KT, Olson WC, Fanous IS, Nail CJ, CM, Hall EH, Slingluff CL Jr. A pilot study 9-peptide plus poly-ICLC early stage cancer. J Immunother Cancer 2017 Nov 21; 5 (1):92. Disclosure Information C. Bogaert: None. L. Van Oudenhove: Pfitzer: N. Mill: B. Fant: