Dynamic applicability domain (dAD): compound-target binding affinity estimates with local conformal prediction.

MOTIVATION: Increasing efforts are being made in the field of machine learning to advance the learning of robust and accurate models from experimentally measured data and enable more efficient drug discovery processes. The prediction of binding affinity is one of the most frequent tasks of compound bioactivity modelling. Learned models for binding affinity prediction are assessed by their average performance on unseen samples, but point predictions are typically not provided with a rigorous confidence assessment. Approaches, such as the conformal predictor framework equip conventional models with a more rigorous assessment of confidence for individual point predictions. In this article, we extend the inductive conformal prediction framework for interaction data, in particular the compound-target binding affinity prediction task. The new framework is based on dynamically defined calibration sets that are specific for each testing pair and provides prediction assessment in the context of calibration pairs from its compound-target neighbourhood, enabling improved estimates based on the local properties of the prediction model. RESULTS: The effectiveness of the approach is benchmarked on several publicly available datasets and tested in realistic use-case scenarios with increasing levels of difficulty on a complex compound-target binding affinity space. We demonstrate that in such scenarios, novel approach combining applicability domain paradigm with conformal prediction framework, produces superior confidence assessment with valid and more informative prediction regions compared to other 'state-of-the-art' conformal prediction approaches. AVAILABILITY AND IMPLEMENTATION: Dataset and the code are available on GitHub (https://github.com/mlkr-rbi/dAD).

cfDNA methylation in liquid biopsies as potential testicular seminoma biomarker.

Background: Seminoma is a testicular tumor type, routinely diagnosed after orchidectomy. As cfDNA represents a source of minimally invasive seminoma patient management, this study aimed to investigate whether cfDNA methylation of six genes from liquid biopsies, have potential as novel seminoma biomarkers. Materials & methods: cfDNA methylation from liquid biopsies was assessed by pyrosequencing and compared with healthy volunteers' samples. Results: Detailed analysis revealed specific CpGs as possible seminoma biomarkers, but receiver operating characteristic curve analysis showed modest diagnostic performance. In an analysis of panels of statistically significant CpGs, two DNA methylation panels emerged as potential seminoma screening panels, one in blood CpG8/CpG9/CpG10 (KITLG) and the other in seminal plasma CpG1(MAGEC2)/CpG1(OCT3/4). Conclusion: The presented data promote the development of liquid biopsy epigenetic biomarkers in the screening of seminoma patients.

Seminoma belongs to testicular cancer, which represents a common malignancy among men of reproductive age. Diagnosis of seminoma is a multistep process that also includes checking tumor biomarkers from blood. However, these biomarkers are not specific for seminoma and to conclude a definite diagnosis of seminoma immunohistochemical analysis is needed, which requires the removal of a whole or partial testicle. Therefore, there is a need for novel, noninvasive biomarkers. cfDNA is the most extensively investigated source of minimally invasive tumor markers. Therefore, this study investigated cfDNA methylation of six genes as potential noninvasive biomarkers for the management of seminoma patients. By examining CpG sites of selected genes by pyrosequencing, the authors detected significant differences. However, receiver operating characteristic curve analysis showed modest results. Therefore, the authors tested possible panels of significantly different CpGs and detected two possible DNA methylation panels for seminoma screening. These findings suggest the further investigation of possible epigenetic biomarkers for seminoma patient management from liquid biopsies.

Comprehensive machine learning based study of the chemical space of herbicides.

Widespread use of herbicides results in the global increase in weed resistance. The rotational use of herbicides according to their modes of action (MoAs) and discovery of novel phytotoxic molecules are the two strategies used against the weed resistance. Herein, Random Forest modeling was used to build predictive models and establish comprehensive characterization of structure-activity relationships underlying herbicide classifications according to their MoAs and weed selectivity. By combining the predictive models with herbicide-likeness rules defined by selected molecular features (numbers of H-bond acceptors and donors, logP, topological and relative polar surface area, and net charge), the virtual stepwise screening platform is proposed for characterization of small weight molecules for their phytotoxic properties. The screening cascade was applied on the data set of phytotoxic natural products. The obtained results may be valuable for refinement of herbicide rotational program as well as for discovery of novel herbicides primarily among natural products as a source for molecules of novel structures and novel modes of action and translocation profiles as compared with the synthetic compounds.

Bioprospecting for Genes Encoding Hydrocarbon-Degrading Enzymes from Metagenomic Samples Isolated from Northern Adriatic Sea Sediments.

Three metagenomic libraries were constructed using surface sediment samples from the northern Adriatic Sea. Two of the samples were taken from a highly polluted and an unpolluted site respectively. The third sample from a polluted site had been enriched using crude oil. The results of the metagenome analyses were incorporated in the REDPET relational database (http://redpet.bioinfo.pbf.hr/REDPET), which was generated using the previously developed MEGGASENSE platform. The database includes taxonomic data to allow the assessment of the biodiversity of metagenomic libraries and a general functional analysis of genes using hidden Markov model (HMM) profiles based on the KEGG database. A set of 22 specialised HMM profiles was developed to detect putative genes for hydrocarbon-degrading enzymes. Use of these profiles showed that the metagenomic library generated after selection on crude oil had enriched genes for aerobic n-alkane degradation. The use of this system for bioprospecting was exemplified using potential alkB and almA genes from this library.