Computational Taxonomy for visualizing and exploring systems with functional response dynamics: soybean's O2-respiration example

Abstract

A collective of 1684 O2-respiration curves of 183 dimensions are collected along soybeans' germinating process with 72 covariate feature-measurements. We develop a computational thematic platform for first discovering and exposing heterogeneous typical-shapes and then extracting readable, visible and explainable pattern information with biological implications when comparing between them. Via existing knowledge on seeds of much smaller sizes, such collective of O2-respiration curves are expected to generally embrace the two-stage or sigmoidal patterns. However, due to Soybean's relatively large surface area with harder to control and retain experimental conditions, much diverse and heterogeneous shapes are highly expected. Shannon entropy is applied to capture potential categorical nature across the 183 time-points, while scaled mutual information is used to evaluate pattern-consistency. Three motifs are determined to partition the temporal time-points into three regions: W1 on [1, 20], W2 on [21, 60] and W3 on [61, 183]. Each curve is partitioned into three segments accordingly. Hierarchical cluster algorithm is applied onto these motif-region-specific curve-segments, respectively. Five clusters of shapes on W1 are selected, four on W2 and four on W3. We computationally build a tree-geometry-based Taxonomic Hierarchy to organize these 80 curve-shapes into 15 typical shapes. Then we compare among these typical shapes to extract and display their differences through block-sustained heatmaps. Each focal block is expressed va a network of covariate features to reveal dominant interacting effects of various orders. Scientific explanations and understanding are to be rested on the collected heatmaps and coupled networks collectively.