Abstract
Heterogeneous cells within a stable state are controlled by gene regulatory networks. Cell-state state transitions are not just gradually and linearly changing phenomena, but rather can be described as nonlinear critical transitions or tipping points (Moris 2016). This abrupt transition between two stable states has been well described using a tipping-point model (Scheffer 2012). When a complex system arrives at a tipping point, there exists fundamental instability in the system. Although this brings a potential risks of unwanted collapse, it also presents an opportunity for positive change through reversal of trajectory. The idea of tipping-point prediction has been previously pursued and implimented, however existing methods each have their limitations. The ‘early-warning’ package in R (http://www.early-warning-signals.org/) is designed for longitudinal data analysis and is applicable to ecosystems, ecology, space, and other fields. However, we found it hard to apply to high-throughput ‘omics’ data. Another published method for biological tipping-point characterization is Dynamic Network Biomarker (DNB) (Chen 2012). This method relies on indexing gene-oscillation per state, and thus is not always feasible for cross-state comparisons and indication of tipping points. Another published method, Index of critical transition (Ic-score), predicts tipping point is limited by its reliance on pre-defined features (Mojtahedi 2016). The purpose of this R package BioTIP is to characterize Biological Tipping-Points from a high-throughput data matrix where rows represent molecular features such as transcripts and columns represent samples or cells in a data matrix. BioTIP implicates one new method and the two aforementioned published methods(DNB and Ic-score) to allow for flexible analysis of not only longitudinal, but also cross-sectional datasets. Besides overcoming the limitations of previous methods, BioTIP provides functions to optimize feature pre-selection and evaluate empirical significance. Additionally, BioTIP defines transcript biotypes according the GENCODE annotation, allowing for study of noncoding transcription (Wang 2018). These improvements allow for an application to cross-sectional datasets. The BioTIP scheme acts as a hybrid model to join the advantages of both DNB and Ic-scoring, and providing enhanced features for high-throughput ‘omics’ data analysis.