The selection of biomarker panels in omics data, challenged by numerous molecular features and limited samples, often requires the use of machine learning methods paired with wrapper feature selection techniques, like genetic algorithms. They test various feature sets-potential biomarker solutions-to fine-tune a machine learning model's performance for supervised tasks, such as classifying cancer subtypes. This optimization process is undertaken using validation sets to evaluate and identify the most effective feature combinations. Evaluations have performance estimation error, measurable as discrepancy between validation and test set performance, and when the selection involves many models the best ones are almost certainly overestimated. This issue is also relevant in a multi-objective feature selection process where various characteristics of the biomarker panels are optimized, such as predictive performances and feature set size. Methods have been proposed to reduce the overestimation after a model has already been selected in single-objective problems, but no algorithm existed capable of reducing the overestimation during the optimization, improving model selection, or applied in the more general multi-objective domain. We propose Dual-stage Optimizer for Systematic overestimation Adjustment in Multi-Objective problems (DOSA-MO), a novel multi-objective optimization wrapper algorithm that learns how the original estimation, its variance, and the feature set size of the solutions predict the overestimation. DOSA-MO adjusts the expectation of the performance during the optimization, improving the composition of the solution set. We verify that DOSA-MO improves the performance of a state-of-the-art genetic algorithm on left-out or external sample sets, when predicting cancer subtypes and/or patient overall survival, using three transcriptomics datasets for kidney and breast cancer.

Machine learning algorithms have been extensively used for accurate classification of cancer subtypes driven by gene expression-based biomarkers. However, biomarker models combining multiple gene expression signatures are often not reproducible in external validation datasets and their feature set size is often not optimized, jeopardizing their translatability into cost-effective clinical tools. We investigated how to solve the multi-objective problem of finding the best trade-offs between classification performance and set size applying seven algorithms for machine learning-driven feature subset selection and analyse how they perform in a benchmark with eight large-scale transcriptome datasets of cancer, covering both training and external validation sets. The benchmark includes evaluation metrics assessing the performance of the individual biomarkers and the solution sets, according to their accuracy, diversity, and stability of the composing genes. Moreover, a new evaluation metric for cross-validation studies is proposed that generalizes the hypervolume, which is commonly used to assess the performance of multi-objective optimization algorithms. Biomarkers exhibiting 0.8 of balanced accuracy on the external dataset for breast, kidney and ovarian cancer using respectively 4, 2 and 7 features, were obtained. Genetic algorithms often provided better performance than other considered algorithms, and the recently proposed NSGA2-CH and NSGA2-CHS were the best performing methods in most cases.

The pharmacological arsenal against the COVID-19 pandemic is largely based on generic anti-inflammatory strategies or poorly scalable solutions. Moreover, as the ongoing vaccination campaign is rolling slower than wished, affordable and effective therapeutics are needed. To this end, there is increasing attention toward computational methods for drug repositioning and de novo drug design. Here, multiple data-driven computational approaches are systematically integrated to perform a virtual screening and prioritize candidate drugs for the treatment of COVID-19. From the list of prioritized drugs, a subset of representative candidates to test in human cells is selected. Two compounds, 7-hydroxystaurosporine and bafetinib, show synergistic antiviral effects in vitro and strongly inhibit viral-induced syncytia formation. Moreover, since existing drug repositioning methods provide limited usable information for de novo drug design, the relevant chemical substructures of the identified drugs are extracted to provide a chemical vocabulary that may help to design new effective drugs.

MOTIVATION: In modern translational research, the development of biomarkers heavily relies on use of omics technologies, but implementations with basic data mining algorithms frequently lead to false positives. Non-dominated Sorting Genetic Algorithm II (NSGA2) is an extremely effective algorithm for biomarker discovery but has been rarely evaluated against large-scale datasets. The exploration of the feature search space is the key to NSGA2 success but in specific cases NSGA2 expresses a shallow exploration of the space of possible feature combinations, possibly leading to models with low predictive performances. RESULTS: We propose two improved NSGA2 algorithms for finding subsets of biomarkers exhibiting different trade-offs between accuracy and feature number. The performances are investigated on gene expression data of breast cancer patients. The results are compared with NSGA2 and LASSO. The benchmarking dataset includes internal and external validation sets. The results show that the proposed algorithms generate a better approximation of the optimal trade-offs between accuracy and set size. Moreover, validation and test accuracies are better than those provided by NSGA2 and LASSO. Remarkably, the GA-based methods provide biomarkers that achieve a very high prediction accuracy (>80%) with a small number of features (<10), representing a valid alternative to known biomarker models, such as Pam50 and MammaPrint. AVAILABILITY AND IMPLEMENTATION: The software is publicly available on GitHub at github.com/UEFBiomedicalInformaticsLab/BIODAI/tree/main/MOO. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Risk assessment is an important decision support task in many domains, including health, engineering, process management, and economy. There is a growing interest in automated methods for risk assessment. These methods should be able to process information efficiently and with little user involvement. Currently, from the scientific literature in the health domain, there is availability of evidence-based knowledge about specific risk factors. On the other hand, there is no automatic procedure to exploit this available knowledge in order to create a general risk assessment tool which can combine the available quantitative data about risk factors and their impact on the corresponding risk. We present a Framework for the Assessment of Risk of adverse Events (FARE) and its first concrete applications FRAT-up and DRAT-up, which were used for fall and depression risk assessment in older persons and validated on four and three European epidemiological datasets, respectively. FARE consists of i) a novel formal ontology called On2Risk; and ii) a logical and probabilistic rule-based model. The ontology was designed to represent qualitative and quantitative data about risks in a general, structured and machine-readable manner so that this data may be concretely exploited by risk assessment algorithms. We describe the structure of the FARE model in the form of logic and probabilistic rules. We show how when starting from machine-readable data about risk factors, like the data contained in On2Risk, an instance of the algorithm can be automatically constructed and used to estimate the risk of an adverse event.

Assessing the risk to develop a specific disease is the first step towards prevention, both at individual and population levels. The development and validation of risk prediction models (RPMs) is the norm within different fields of medicine but still underused in psychiatry, despite the global impact of mental disorders. In particular, there is a lack of RPMs to assess the risk of developing depression, the first worldwide cause of disability and harbinger of functional decline in old age. We present the depression risk assessment tool DRAT-up, the first prospective RPM to identify late-life depression among community-dwelling subjects aged 60-75. The development of DRAT-up was based on appraisal of relevant literature, extraction of robust risk estimates, and integration into model parameters. A unique feature is the ability to estimate risk even in the presence of missing values. To assess the properties of DRAT-up, a validation study was conducted on three European cohorts, namely, the English Longitudinal Study of Ageing, the Invecchiare nel Chianti, and the Irish Longitudinal Study on Ageing, with 20 206, 1359, and 3124 eligible samples, respectively. The model yielded accurate risk estimation in the three datasets from a small number of predictors. The Brier scores were 0.054, 0.133, and 0.041, respectively, while the values of area under the curve (AUC) were 0.761, 0.736, and 0.768, respectively. Sensitivity analyses suggest robustness to missing values: setting any individual feature to unknown caused the Brier scores to increase by 0.004 and the AUCs to decrease by 0.045 in the worst cases. DRAT-up can be readily used for clinical purposes and to aid policy-making in the field of mental health.

Fall risk assessment is important for fall prediction and fall prevention among older adults. In order to be applied in clinical practice, a fall risk assessment tool need to be regarded useful. The aim of this study was to evaluate healthcare professionals' user experience (UX) of the Fall Risk Assessment Tool (FRAT-up). Eight health care professionals working in a geriatric hospital department or in the primary health care system participated in a focus group and evaluated the tool. The healthcare professionals considered the FRAT-up tool to be novel and considered it useful in clinical practice. The tool can be used to get an extended knowledge and understanding of how to individually tailor a fall prevention intervention. It is suggested that for a better experience, the scale should be simplified and the tool should be integrated with the patient's medical record.

About one-third of persons over 65 are subject to at least one fall during a year, and many of them are subjected to health, psychological and financial consequences. A requirement to improve the effectiveness of preventive interventions is to timely identify subjects at higher risk. In this work we introduce the Farseeing Fall Risk Assessment Tool (FRAT-up), a software tool for evaluating the fall risk of a subject, based on known risk factors. The tool is based on probabilistic rules, generated automatically from a light ontology capturing the scientific findings about risk factors. FRAT-up has been tested on the In CHIANTI dataset, showing performances comparable with state-of-the-art tools.