(516c) Predicting Colorectal Cancer from Human Gut Microbiome Data

Dysbiosis in the human gut microbiome has been associated with colorectal cancer (CRC) and using microbiome data as a diagnostic tool has been a subject of extensive research in the past decade. From the microbiome composition of 16 subjects and 69 total replicates, this study develops a binary logistic regression (BLR) model that can identify significant phyla of microbes with high predictive power for subjects who do not have CRC. Preliminary results show that the abundance of Actinobacteria, Tenericutes, Verrucomicrobia, along with Age and BMI are factors that are most predictive of the presence of CRC. In addition, this study sought for phyla that can be used to diagnose subjects without CRC more precisely, for the purpose of reducing unnecessary colonoscopies, and to screen for potentially afflicted patients.

Predicted probabilities obtained from the BLR for non-CRC patients were used as the indicator of the predictive power of the model. The lower the value of this indicator, the better the model is at classifying those subjects without CRC. Using a cutoff at 5% for the indicator, Actinobacteria, Bacteroidetes, Tenericutes and BMI have been identified as potential parameters to use to differentiate those without CRC from those with CRC. This framework is applicable to other diseases (such as cystic fibrosis) with potential to become a useful screening tool in the healthcare field.