Alex Rubinsteyn

Using a Machine Learning Approach to Predict Outcomes after Radiosurgery for Cerebral Arteriovenous Malformations

Predictions of patient outcomes after a given therapy are fundamental to medical practice. We employ a machine learning approach towards predicting the outcomes after stereotactic radiosurgery for cerebral arteriovenous malformations (AVMs). Using three prospective databases, a machine learning approach of feature engineering and model optimization was implemented to create the most accurate predictor of AVM outcomes. Existing prognostic systems were scored for purposes of comparison. The final predictor was secondarily validated on an independent site’s dataset not utilized for initial construction. Out of 1,810 patients, 1,674 to 1,291 patients depending upon time threshold, with 23 features were included for analysis and divided into training and validation sets. The best predictor had an average area under the curve (AUC) of 0.71 compared to existing clinical systems of 0.63 across all time points. On the heldout dataset, the predictor had an accuracy of around 0.74 at across all time thresholds with a specificity and sensitivity of 62% and 85% respectively. This machine learning approach was able to provide the best possible predictions of AVM radiosurgery outcomes of any method to date, identify a novel radiobiological feature (3D surface dose), and demonstrate a paradigm for further development of prognostic tools in medical care.

Predicting Peptide-MHC Binding Affinities With Imputed Training Data

Predicting the binding affinity between MHC proteins and their peptide ligands is a key problem in computational immunology. State of the art performance is currently achieved by the allele-specific predictor NetMHC and the pan-allele predictor NetMHCpan, both of which are ensembles of shallow neural networks. We explore an intermediate between allele-specific and pan-allele prediction: training allele-specific predictors with synthetic samples generated by imputation of the peptide-MHC affinity matrix. We find that the imputation strategy is useful on alleles with very little training data. We have implemented our predictor as an open-source software package called MHCflurry and show that MHCflurry achieves competitive performance to NetMHC and NetMHCpan.

MHCflurry: open-source class I MHC binding affinity prediction

Predicting the binding affinity of major histocompatibility complex I (MHC I) proteins and their peptide ligands is important for vaccine design. We introduce an open-source package for MHC I binding prediction, MHCflurry. The software implements allele-specific neural networks that use a novel architecture and peptide encoding scheme. When trained on affinity measurements, MHCflurry outperformed the standard predictors NetMHC 4.0 and NetMHCpan 3.0 overall and particularly on non-9-mer peptides in a benchmark of ligands identified by mass spectrometry. The released predictor, MHCflurry 1.2.0, uses mass spectrometry datasets for model selection and showed competitive accuracy with standard tools, including the recently released NetMHCpan 4.0, on a small benchmark of affinity measurements. MHCflurrys prediction speed exceeded 7,000 predictions per second, 396 times faster than NetMHCpan 4.0. MHCflurry is freely available to use, retrain, or extend, includes Python library and command line interfaces, may be installed using package managers, and applies software development best practices.

Vaxrank: A Computational Tool For Designing Personalized Cancer Vaccines

Therapeutic vaccines targeting mutant tumor antigens (“neoantigens”) are an increasingly popular form of personalized cancer immunotherapy. Vaxrank is a computational tool for selecting neoantigen vaccine peptides from tumor mutations, tumor RNA data, and patient HLA type. Vaxrank is freely available at www.github.com/hammerlab/vaxrank under the Apache 2.0 open source license and can also be installed from the Python Package Index.

Personalized neoantigen vaccination with synthetic long peptides

§ All non-synonymous somatic variants identified via WES translated into a 40+ character long string corresponding to the amino acid sequence of the mutated genetic region. § Every string is broken into multiple 8-11 character long overlapping substrings, each of which is assessed for pMHC binding affinity, and immunogenicity using NetMHCcons. § Mutated amino acid sequences are ranked based on the sum of putative epitope scores contained in their sequence, and 31-mer vaccine peptides are chosen from the twenty highest scoring sequences via sliding window optimization. § Among all 31-mer sliding windows with equivalent mutant epitope content (candidate vaccine peptides), we select the vaccine peptide with the least number of wildtype epitopes. § The mutation in each vaccine peptide must be at least 5 residues from either the beginning or end of the peptide sequence. References

Neoantigen homology and predicting response to immune checkpoint blockade in cancer

● Pathogen Homology ○ Predicted neoantigens were aligned with T-cell positive peptides from IEDB of the same length, considering positions 3 through n-1 (n = length) ○ Peptide alignment was scored with the PMBEC matrix [3] and a gap penalty of min(PMBEC). For example, the following entry had a score of 1.4: Immune checkpoint inhibitors are promising cancer treatments for a variety of malignancies, but accurate prediction of clinical response remains an active area of research.