Biology subjects

Konopka, T.

Publications and source records attributed to Konopka, T..

5 recordsLinked to original sources

bioRxiv2021

Correcting machine learning models using calibrated ensembles with "mlensemble"

Machine learning models in bioinformatics are often trained and used within the scope of a single project, but some models are also reused across projects and deployed in translational settings. Over time, trained models may turn out to be maladjusted to the properties of new data. This creates the need to improve their performance under various constraints. This work explores correcting models without retraining from scratch and without accessing the original training data. It uses a taxonomy of strategies to guide the development of a software package, mlensemble. Key features include joining heterogeneous models into ensembles and calibrating ensembles to the properties of new data. These are well-established techniques but are often hidden within more complex tools. By exposing them to the application level, the package enables analysts to use expert knowledge to adjust models whenever needed. Calculations with imaging data show benefits when the noise characteristics of the training and the application datasets differ. An example using genomic single-cell data demonstrates model portability despite batch effects. The generality of the framework makes it applicable also in other subject domains.

bioinformatics↗

bioRxiv2021

Dimensional reduction of phenotypes from 53,000 mouse models reveals a diverse landscape of gene function

Animal models have long been used to study gene function and the impact of genetic mutations on phenotype. Through the research efforts of thousands of research groups, systematic curation of published literature, and high-throughput phenotyping screens, the collective body of knowledge for the mouse now covers the majority of protein-coding genes. We here collected data for over 53,000 mouse models with mutations in over 15,000 genomic markers and characterized by more than 254,000 annotations using more than 9,000 distinct ontology terms. We investigated dimensional reduction and embedding techniques as means to facilitate access to this diverse and high-dimensional information. Our analyses provide the first visual maps of the landscape of mouse phenotypic diversity. We also summarize some of the difficulties in producing and interpreting embeddings of sparse phenotypic data. In particular, we show that data preprocessing, filtering, and encoding have as much impact on the final embeddings as the process of dimensional reduction. Nonetheless, techniques developed in the context of dimensional reduction create opportunities for explorative analysis of this large pool of public data, including for searching for mouse models suited to study human diseases.

bioinformatics↗

bioRxiv2021

A pan-ontology view of machine-derived knowledge representations and feedback mechanisms for curation

Biomedical ontologies are established tools that organize knowledge in specialized research areas. They can also be used to train machine-learning models. However, it is unclear to what extent representations of ontology concepts learned by machine-learning models capture the relationships intended by ontology curators. It is also unclear whether the representations can provide insights to improve the curation process. Here, we investigate ontologies from across the spectrum of biological research and assess the concordance of formal ontology hierarchies with representations based on plain-text definitions. By comparing the internal properties of each ontology, we describe general patterns across the pan-ontology landscape and pinpoint areas with discrepancies in individual domains. We suggest specific mechanisms through which machine-learning approaches can lead to clarifications of ontology definitions. Synchronizing patterns in machine-derived representations with those intended by the ontology curators will likely streamline the use of ontologies in downstream applications.

bioinformatics↗

bioRxiv2020

A pan-cancer landscape of somatic substitutions in non-unique regions of the human genome

Around 13% of the human genome displays high sequence similarity with at least one other chromosomal position and thereby poses challenges for computational analyses such as detection of somatic events in cancer. We here extract features of sequencing data from across non-unique regions and employ a machine learning pipeline to describe a landscape of somatic substitutions in 2,658 cancers from the PCAWG cohort. We show mutations in non-unique regions are consistent with mutations in unique regions in terms of mutation load and substitution profiles, and can be validated with linked-read sequencing. This uncovers hidden mutations in ~1,700 coding sequences and thousands of regulatory elements, including known cancer genes, immunoglobulins, and highly mutated gene families.

genomics↗

bioRxiv2019

Human and mouse essentiality screens as a resource for disease gene discovery

Although genomic sequencing has been transformative in the study of rare genetic diseases, identifying causal variants remains a considerable challenge that can be addressed in part by new gene-specific knowledge. Here, we integrate measures of how essential a gene is to supporting life, as inferred from the comprehensive viability and phenotyping screens performed on knockout mice by the International Mouse Phenotyping Consortium and from human cell line essentiality screens. We propose a novel, cross-species gene classification across the Full Spectrum of Intolerance to Loss-of-function (FUSIL) and demonstrate that genes in five mutually exclusive FUSIL categories have differing characteristics in the biological processes they regulate, tissue expression levels and human mutation rates. Most notably, Mendelian disease genes, particularly those associated with developmental disorders, are highly overrepresented in the developmental lethal category, representing genes not essential for cell survival but required for organism development. Exploiting this finding, we have screened developmental disorder cases from three independent disease sequencing consortia and identified potentially pathogenic, de novo variants shared in different patients for several developmental lethal genes that have not previously been associated with rare disease. We therefore propose FUSIL as an efficient resource for disease gene discovery.

genomics↗