Search

Precision medicine is empowered by biomarker discovery. High-throughput drug screens in cancer cell lines enable the identification of drug sensitivity biomarkers, whilst resistance biomarkers remain underexplored: dissecting meaningful drug responses at high concentrations is impossible due to cytotoxicity, making non-responders and resistant cell lines indistinguishable, thus limiting resistance biomarker discovery to frequently mutated cancer genes. In order to address this, we developed a statistical method to indirectly identify resistance biomarkers. Here, we interrogate subpopulations carrying sensitivity biomarkers, and consecutively investigate resistant outliers for unique genetic alterations that putatively drive this resistance. Out of 78 cases with resistant outliers, our method predicts infrequent resistance biomarkers for 43 of them, e.g. the clinically relevant EGFRT790M mutation in lung adenocarcinoma treated with EGFR inhibitors. Interrogating the underpinnings of drug resistance with publicly available CRISPR phenotypic assays assists in prioritising resistance drivers, offering hypotheses for potentially effective drug combinations, and thereby paving the way for the next generation of precision medicine.

This paper presents a novel neural learning algorithm for analysing protein peptides which comprise amino acids as non-numerical attributes. The algorithm is derived from the radial basis function neural networks (RBFNNs) and is referred to as a bio-basis function neural network (BBFNN). The basic principle is to replace the radial basis function used by RBFNNs with a bio-basis function. Each basis in BBFNN is supported by a peptide. The bases collectively form a feature space, in which each basis represents a feature dimension. A linear classifier is constructed in the feature space for characterising a protein peptide in terms of functional status. The theoretical basis of BBFNN is that peptides, which perform the same function will have similar compositions of amino acids. Because of this, the similarity between peptides can have statistical significance for modelling while the proposed bio-basis function can well code this information from data. The application to two real cases shows that BBFNN outperformed multi-layer perceptrons and support vector machines.

This paper presents an algorithm which is able to extract discriminant rules from oligopeptides for protease proteolytic cleavage activity prediction. The algorithm is developed using genetic programming. Three important components in the algorithm are a min-max scoring function, the reverse Polish notation (RPN) and the use of minimum description length. The min-max scoring function is developed using amino acid similarity matrices for measuring the similarity between an oligopeptide and a rule, which is a complex algebraic equation of amino acids rather than a simple pattern sequence. The Fisher ratio is then calculated on the scoring values using the class label associated with the oligopeptides. The discriminant ability of each rule can therefore be evaluated. The use of RPN makes the evolutionary operations simpler and therefore reduces the computational cost. To prevent overfitting, the concept of minimum description length is used to penalize over-complicated rules. A fitness function is therefore composed of the Fisher ratio and the use of minimum description length for an efficient evolutionary process. In the application to four protease datasets (Trypsin, Factor Xa, Hepatitis C Virus and HIV protease cleavage site prediction), our algorithm is superior to C5, a conventional method for deriving decision trees.