Biology subjects

Stoll, P.

Publications and source records attributed to Stoll, P..

4 recordsLinked to original sources

bioRxiv2021

Interpretable machine learning identifies paediatric Systemic Lupus Erythematosus subtypes based on gene expression data

Transcriptomic analyses are commonly used to identify differentially expressed genes between patients and controls, or within individuals across disease courses. These methods, whilst effective, cannot encompass the combinatorial effects of genes driving disease. We applied rule-based machine learning (RBML) models and rule networks (RN) to an existing paediatric Systemic Lupus Erythematosus (SLE) blood expression dataset, with the goal of developing gene networks to separate low and high disease activity (DA1 and DA3). The resultant model had an 81% accuracy to distinguish between DA1 and DA3, with unsupervised hierarchical clustering revealing additional subgroups indicative of the immune axis involved or state of disease flare. These subgroups correlated with clinical variables, suggesting that the gene sets identified may further the understanding of gene networks that act in concert to drive disease progression. This included roles for genes i) induced by interferons (IFI35 and OTOF), ii) key to SLE cell types (KLRB1 encoding CD161), or iii) with roles in autophagy and NF-{kappa}B pathway responses (CKAP4). As demonstrated here, RBML approaches have the potential to reveal novel gene patterns from within a heterogeneous disease, facilitating patient clinical and therapeutic stratification.

bioinformatics↗

bioRxiv2020

Accuracy of predicting chemical body composition of growing pigs using dual-energy X-ray absorptiometry

AO_SCPLOWBSTRACTC_SCPLOWStudies in animal science assessing nutrient and energy efficiency or determining nutrient requirements necessitate gathering exact measurements of body composition or body nutrient contents. Wet chemical analysis methods or standardized dissection are commonly applied, but both are destructive. Harnessing human medical imaging techniques for animal science can enable repeated measurements of individuals over time and reduce the number of individuals required for research. Among imaging techniques, dual-energy X-ray absorptiometry (DXA) is particularly promising. However, the measurements obtained with DXA do not perfectly match dissections or chemical analyses, requiring the adjustment of the DXA via calibration equations. Several calibration regressions have been published, but comparative studies are pending. Thus, it is currently not clear whether existing regression equations can be directly used to convert DXA measurements into chemical values or whether each individual DXA device will require its own calibration. Our study builds prediction equations that relate body composition to the content of single nutrients in growing entire male pigs (body weight range 20-100 kg) as determined by both DXA and chemical analyses, with R2 ranging between 0.89 for ash and 0.99 for water and crude protein. Moreover, we show that the chemical composition of the empty body can be satisfactorily determined by DXA scans of carcasses, with the prediction error rCV ranging between 4.3% for crude protein and 12.6% for ash. Finally, we compare existing prediction equations for pigs of a similar range of body weights with the equations derived from our DXA measurements and evaluate their fit with our chemical analyses data. We found that existing equations for absolute contents that were built using the same DXA beam technology predicted our data more precisely than equations based on different technologies and percentages of fat and lean mass. This indicates that the creation of generic regression equations that yield reliable estimates of body composition in pigs of different growth stages, sexes and genetic breeds could be achievable in the near future. DXA may be a promising tool for high-throughput phenotyping for genetic studies, because it efficiently measures body composition in a large number and wide array of animals.

physiology↗

bioRxiv2020

Including tree spatial extension in the evaluation of neighbourhood competition effects in Bornean rain forest

Classical tree neighbourhood models use size variables acting at point distances. In a new approach here, trees were spatially extended as a function of their crown sizes, represented impressionistically as points within crown areas. Extension was accompanied by plasticity in the form of crown removal or relocation under the overlap of taller trees. Root systems were supposedly extended in a similar manner. For the 38 most abundant species in the focal size class (10 - <100 cm stem girth) in two 4-ha plots at Danum (Sabah), for periods P1 (1986-1996) and P2 (1996-2007), stem growth rate and tree survival were individually regressed against stem size, and neighbourhood conspecific (CON) and heterospecific (HET) basal areas within incremented steps in radius. Model parameters were critically assessed, and statistical robustness in the modelling set by randomization testing. Classical and extended models differed importantly in their outcomes. Crown extension weakened the relationship of CON effect on growth versus plot species abundance, showing that models without plasticity over-estimated negative density dependence. A significant negative trend of difference in CON effects on growth (P2 - P1) versus CON or HET effect on survival in P1 was strongest with crown extension. Model outcomes did not then support an explanation of CON and HET effects being due to (asymmetric) competition for light alone. An alternative hypothesis is that changes in CON effects on small trees, largely incurred by a drought phase (relaxing light limitation) in P2, and following the more shaded (suppressing) conditions in P1, were likely due to species-specific (symmetric) root competition and mycorrhizal processes. The very high variation in neighbourhood composition and abundances led to a strong neighbourhood stochasticity, and hence to largely idiosyncratic species responses. A need to much better understand the roles of rooting structure and processes at the individual tree level was highlighted.

ecology↗

bioRxiv2019

R.ROSETTA: a package for analysis of rule-based classification models

MotivationFor machine learning to matter beyond intellectual curiosity, the models developed therefrom must be adopted within the greater scientific community. In this study, we developed an interpretable machine learning framework that allows identification of semantics from various datatypes. Our package can analyze and illuminate co-predictive mechanisms reflecting biological processes. ResultsWe present R.ROSETTA, an R package for building and analyzing interpretable machine learning models. R.ROSETTA gathers combinatorial statistics via rule-based modelling for accessible and transparent results, well-suited for adoption within the greater scientific community. The package also provides statistics and visualization tools that facilitate minimization of analysis bias and noise. Investigating case-control studies of autism, we showed that our tool provided hypotheses for potential interdependencies among features that discerned phenotype classes. These interdependencies regarded neurodevelopmental and autism-related genes. Although our sample application of R.ROSETTA was used for transcriptomic data analysis, R.ROSETTA works perfectly with any decision-related omics data. AvailabilityThe R.ROSETTA package is freely available at https://github.com/komorowskilab/R.ROSETTA. Contactmateusz.garbulowski@icm.uu.se (Mateusz Garbulowski), jan.komorowski@icm.uu.se (Jan Komorowski)

bioinformatics↗