WNT/beta-catenin sinyal yolağında makine öğrenmesi ile hedef genler belirlenmesi

The Wnt signalling pathway is a driving force of proliferation and differentiation. Aberrant behaviour in this pathway may lead to several types of cancers and Alzheimer's disease. This pathway controls the transcription of target genes via modulating the presence of Betacatenin in cytosol, which triggers the TCF/LEF transcription factor. We found 93 target genes identified experimentally in the colorectal cancer context, and new target genes are constantly being discovered. This study aims to identify novel target genes of the Wnt/Beta-catenin signalling pathway using a machine learning approach. We analysed several publicly available Microarray and RNA-seq experiments and used the differential gene expression data to represent the genes. We used the experimentally validated target genes as "positive" examples in training. We chose the "negative" examples randomly from the rest of the genes. We trained pools of 1000 independent classifiers using the Classification and Regression Tree (CART) method. Then each trained classifier was used to assign a "positive" or "negative" label for each gene. The number of times each gene is classified as "positive" is a score that can be tested using the Fisher method. Thus, we found a set of putative target genes having an expression pattern very similar to known target genes. The pool of trained classifiers predicted 144 putative novel target genes. Some of the highest scoring genes are PTCH1, GLI3 and SOX4. The first two predictions, PTCH1 and GLI3, are important components of the Hedgehog Signalling. This suggests a possible interplay between Wnt and Hedgehog signalling in colorectal cancer. In parallel to our study, experimental researchers have reported that PTCH1 is a colon specific Wnt target. We present a bioinformatic method that can be used to predict target genes of the canonical Wnt signalling pathway, and eventually other pathways, based only on gene expression data and a sample of experimentally validated targets. This method narrows the set of genes that should be experimentally validated. Moreover, some of our predictions have already been validated by other studies.