Regression plane concept for analysing continuous cellular processes with machine learning

Abel Szkalisity,Abel Szkalisity,Filippo Piccinini,Filippo Piccinini,Istvan Gergely Varga,Istvan Gergely Varga,Attila Beleon,Elina Ikonen,Istvan Ando,Csaba Molnar,Sanna Timonen,Sanna Timonen,Yohei Yamauchi,Viktor Honti,Viktor Honti,Lassi Paavolainen,Lassi Paavolainen,Peter Horvath,Tamas Balassa,Jaakko Peltonen,Jaakko Peltonen,Vilja Pietiäinen,Vilja Pietiäinen,Indranil Banerjee,Ede Migh

doi:10.1038/s41467-021-22866-x

Abstract

Biological processes are inherently continuous, and the chance of phenotypic discovery is significantly restricted by discretising them. Using multi-parametric active regression we introduce the Regression Plane (RP), a user-friendly discovery tool enabling class-free phenotypic supervised machine learning, to describe and explore biological data in a continuous manner. First, we compare traditional classification with regression in a simulated experimental setup. Second, we use our framework to identify genes involved in regulating triglyceride levels in human cells. Subsequently, we analyse a time-lapse dataset on mitosis to demonstrate that the proposed methodology is capable of modelling complex processes at infinite resolution. Finally, we show that hemocyte differentiation in Drosophila melanogaster has continuous characteristics.

Highlights

Biological processes are inherently continuous, and the chance of phenotypic discovery is significantly restricted by discretising them
A great variety of software tools are available for performing imaging assays in a supervised manner (e.g. CellProfiler Analyst, Ilastik, CellCognition, Advanced Cell Classifier5), all of them rely on the assumption that the underlying biological processes have stable steady states that can be dissected into discrete phenotypic classes (Fig. 1a)
Regression plane is implemented as an opensource module of Advanced Cell Classifier (ACC)[6], and it has been available since ACC v3.0

Summary

Results

We evaluated whether siRNA perturbations of candidate genes, previously revealed to influence blood triglyceride (TG) levels in humans in a genome-wide association study[17], would affect the morphology of lipid droplets (LDs) in cultured hepatocytes (Huh[7] cell line) Regarding their continuous changes in localization, number and size, LDs form a heterogeneous population reflecting different cellular metabolic states[18]. In the case of lamellocytes instead of identifying 2 clearly separated subtypes I and II, we have observed that the differentiation processes are evenly distributed on the regression plane, as reflected by specific features (Fig. 4b, c, f) This finding suggests that type I and type II lamellocytes, both differentiating from plasmatocytes, are not definitely distinguishable cell types, but rather they are two extreme stages of a size continuum (Fig. 4e).

Discussion

Methods

Code availability