Optimization-based predictive models in medicine and biology

Advancements in medical technology have created numerous large datasets including many features. Usually, all captured features are not necessary, and there are redundant and irrelevant features, which reduce the performance of algorithms. To tackle this challenge, many metaheuristic algorithms are used to select effective features. However, most of them are not effective and scalable enough to select effective features from large medical datasets as well as small ones. Therefore, in this paper, a binary moth-flame optimization (B-MFO) is proposed to select effective features from small and large medical datasets. Three categories of B-MFO were developed using S-shaped, V-shaped, and U-shaped transfer functions to convert the canonical MFO from continuous to binary. These categories of B-MFO were evaluated on seven medical datasets and the results were compared with four well-known binary metaheuristic optimization algorithms: BPSO, bGWO, BDA, and BSSA. In addition, the convergence behavior of the B-MFO and comparative algorithms were assessed, and the results were statistically analyzed using the Friedman test. The experimental results demonstrate a superior performance of B-MFO in solving the feature selection problem for different medical datasets compared to other comparative algorithms.

A prediction model including age, self-rated health (SRH) and prior sickness absence (SA) has previously been found to predict frequent SA. To further validate the model and develop it for clinical use. A multicentre study of care of the elderly workers employed at one of 14 centres in Aarhus (Denmark). SA episodes recorded in the year prior to baseline and both age and SRH at baseline were included in a prediction model for frequent (three or more) SA episodes during a 1-year follow-up period. The prediction model was developed in the largest centre. Risk predictions and discrimination between high- and low-risk workers were investigated in the other centres. The prediction rule 'SRH-prior SA' was derived from the prediction model and prognostic properties of the prediction rule were investigated for each centre, using score <0 as cut-off. Of 2562 workers, 1930 had complete data for analysis. Predictions were accurate in 4 of 13 centres; discrimination was good in five and fair in another five centres. Prediction rule scores <0 identified workers at risk of frequent SA with sensitivities of 0.17-0.54, specificities of 0.86-0.96 and positive predictive values of 0.54-0.87 across centres. The prediction model discriminated between workers at high and low risk of frequent SA in the majority of centres. The prediction rule 'SRH-prior SA' can be used in clinical practice specifically to identify workers at high risk of frequent SA.

Background and purpose — External validation of machine learning (ML) prediction models is an essential step before clinical application. We assessed the proportion, performance, and transparent reporting of externally validated ML prediction models in orthopedic surgery, using the Transparent Reporting for Individual Prognosis or Diagnosis (TRIPOD) guidelines. Material and methods — We performed a systematic search using synonyms for every orthopedic specialty, ML, and external validation. The proportion was determined by using 59 ML prediction models with only internal validation in orthopedic surgical outcome published up until June 18, 2020, previously identified by our group. Model performance was evaluated using discrimination, calibration, and decision-curve analysis. The TRIPOD guidelines assessed transparent reporting. Results — We included 18 studies externally validating 10 different ML prediction models of the 59 available ML models after screening 4,682 studies. All external validations identified in this review retained good discrimination. Other key performance measures were provided in only 3 studies, rendering overall performance evaluation difficult. The overall median TRIPOD completeness was 61% (IQR 43–89), with 6 items being reported in less than 4/18 of the studies. Interpretation — Most current predictive ML models are not externally validated. The 18 available external validation studies were characterized by incomplete reporting of performance measures, limiting a transparent examination of model performance. Further prospective studies are needed to validate or refute the myriad of predictive ML models in orthopedics while adhering to existing guidelines. This ensures clinicians can take full advantage of validated and clinically implementable ML decision tools.

The abundance of medical resources has encouraged the development of systems that allow for efficient searches of information in large medical image data sets. State‐of‐the‐art image retrieval models are classified into three categories: content‐based (visual) models, textual models, and combined models. Content‐based models use visual features to answer image queries, textual image retrieval models use word matching to answer textual queries, and combined image retrieval models, use both textual and visual features to answer queries. Nevertheless, most of previous works in this field have used the same image retrieval model independently of the query type. In this article, we define a list of generic and specific medical query features and exploit them in an association rule mining technique to discover correlations between query features and image retrieval models. Based on these rules, we propose to use an associative classifier (NaiveClass) to find the best suitable retrieval model given a new textual query. We also propose a second associative classifier (SmartClass) to select the most appropriate default class for the query. Experiments are performed on Medical ImageCLEF queries from 2008 to 2012 to evaluate the impact of the proposed query features on the classification performance. The results show that combining our proposed specific and generic query features is effective in query classification.

FRI0062 A prediction rule for the development of arthritis in seropositive arthralgia patients

This dissertation contributes with novel predictive modelling approaches to data-driven computational psychiatry and offers alternative analyses frameworks to the standard statistical analyses in psychiatric research. In particular, this document advances research in medical data mining, especially psychiatry, via two phases. In the first phase, this document promotes research by proposing synergistic machine learning and statistical approaches for detecting patterns and developing predictive models in clinical psychiatry data to classify diseases, predict treatment outcomes or improve treatment selections. In particular, these data-driven approaches are built upon several machine learning techniques whose predictive models have been pre-processed, trained, optimised, post-processed and tested in novel computationally intensive frameworks. In the second phase, this document advances research in medical data mining by proposing several novel extensions in the area of data classification by offering a novel decision tree algorithm, which we call PIDT, based on parameterised impurities and statistical pruning approaches toward building more accurate decision trees classifiers and developing new ensemblebased classification methods. In particular, the experimental results show that by building predictive models with the novel PIDT algorithm, these models primarily led to better performance regarding accuracy and tree size than those built with traditional decision trees. The contributions of the proposed dissertation can be summarised as follow. Firstly, several statistical and machine learning algorithms, plus techniques to improve these algorithms, are explored. Secondly, prediction modelling and pattern detection approaches for the first-episode psychosis associated with cannabis use are developed. Thirdly, a new computationally intensive machine learning framework for understanding the link between cannabis use and first-episode psychosis was introduced. Then, complementary and equally sophisticated prediction models for the first-episode psychosis associated with cannabis use were developed using artificial neural networks and deep learning within the proposed novel computationally intensive framework. Lastly, an efficient novel decision tree algorithm (PIDT) based on novel parameterised impurities and statistical pruning approaches is proposed and tested with several medical datasets. These contributions can be used to guide future theory, experiment, and treatment development in medical data mining, especially psychiatry.

Machine learning methods are used to discover complex nonlinear relationships in biological and medical data. However, sophisticated learning models are computationally unfeasible for data with millions of features. Here we introduce the first feature selection method for nonlinear learning problems that can scale up to large, ultra-high dimensional biological data. More specifically, we scale up the novel Hilbert-Schmidt Independence Criterion Lasso (HSIC Lasso) to handle millions of features with tens of thousand samples. The proposed method is guaranteed to find an optimal subset of maximally predictive features with minimal redundancy, yielding higher predictive power and improved interpretability. Its effectiveness is demonstrated through applications to classify phenotypes based on module expression in human prostate cancer patients and to detect enzymes among protein structures. We achieve high accuracy with as few as 20 out of one million features --- a dimensionality reduction of 99.998%. Our algorithm can be implemented on commodity cloud computing platforms. The dramatic reduction of features may lead to the ubiquitous deployment of sophisticated prediction models in mobile health care applications.

Graph neural networks (GNNs) have achieved extraordinary enhancements in various areas including the fields medical imaging and network neuroscience where they displayed a high accuracy in diagnosing challenging neurological disorders such as autism. In the face of medical data scarcity and high-privacy, training such data-hungry models remains challenging. Federated learning brings an efficient solution to this issue by allowing to train models on multiple datasets, collected independently by different hospitals, in fully data-preserving manner. Although both state-of-the-art GNNs and federated learning techniques focus on boosting classification accuracy, they overlook a critical unsolved problem: investigating the reproducibility of the most discriminative biomarkers (i.e., features) selected by the GNN models within a federated learning paradigm. Quantifying the reproducibility of a predictive medical model against perturbations of training and testing data distributions presents one of the biggest hurdles to overcome in developing translational clinical applications. To the best of our knowledge, this presents the first work investigating the reproducibility of federated GNN models with application to classifying medical imaging and brain connectivity datasets. We evaluated our framework using various GNN models trained on medical imaging and connectomic datasets. More importantly, we showed that federated learning boosts both the accuracy and reproducibility of GNN models in such medical learning tasks. Our source code is available at https://github.com/basiralab/reproducibleFedGNN.KeywordsGraph neural networksFederated LearningReproducibilityBrain connectivity graphsPredictive medicine

Modern biological research requires rapid, complex, and reproducible integration of multiple experimental results generated both internally and externally (e.g., from public repositories). Although large systematic meta-analyses are among the most effective approaches both for clinical biomarker discovery and for computational inference of biomolecular mechanisms, identifying, acquiring, and integrating relevant experimental results from multiple sources for a given study can be time-consuming and error-prone. To enable efficient and reproducible integration of diverse experimental results, we developed a novel approach for standardized acquisition and analysis of high-throughput and heterogeneous biological data. This allowed, first, novel biomolecular network reconstruction in human prostate cancer, which correctly recovered and extended the NFκB signaling pathway. Next, we investigated host-microbiome interactions. In less than an hour of analysis time, the system retrieved data and integrated six germ-free murine intestinal gene expression datasets to identify the genes most influenced by the gut microbiota, which comprised a set of immune-response and carbohydrate metabolism processes. Finally, we constructed integrated functional interaction networks to compare connectivity of peptide secretion pathways in the model organisms Escherichia coli, Bacillus subtilis, and Pseudomonas aeruginosa.

Artificial intelligence (AI) has emerged as a revolutionary tool in hematology, capable of analyzing, interpreting, and integrating complex biological, morphological, and genomic data with enhanced precision and speed. This review aims to explore the applications of AI in hematology, with a focus on diagnostic automation, flow cytometry, genomic analysis, predictive modeling, and clinical decision support. A narrative literature review was conducted using PubMed, ScienceDirect, and Google Scholar, covering publications from 2018 to 2024. Keywords included: artificial intelligence, machine learning, deep learning, hematology, cytology, genomics, blood cancers. Articles presenting clinically validated or promising applications were selected. The findings indicate that AI is particularly applied in automated diagnostics and morphological analysis, especially in blood cell classification and the detection of morphological abnormalities. It also plays a pivotal role in flow cytometry and genomics, notably through the interpretation of next-generation sequencing (NGS) data to identify mutations and clinically relevant risk profiles, particularly in acute myeloid leukemia (AML). Furthermore, AI contributes to predictive modeling, patient follow-up, and morphometric quantitative image analysis, facilitating the early detection of hematologic malignancies. In conclusion, AI represents a major technological breakthrough in hematology, complementing human expertise while reducing workload. Future efforts should focus on enhancing algorithm explainability and clinical Regulatory harmonization and close collaboration between clinicians, data scientists, and institutions are essential to fully harness the potential of AI in the diagnosis, prognosis, and treatment of blood disorders.

Synthetic lethality (SL) refers to a genetic interaction in which the simultaneous perturbation of two genes leads to cell or organism death, whereas viability is maintained when only one of the pair is altered. The experimental exploration of these pairs and predictive modeling in computational biology contribute to our understanding of cancer biology and the development of cancer therapies. We extensively reviewed experimental technologies, public data sources, and predictive models in the study of synthetic lethal gene pairs and herein detail biological assumptions, experimental data, statistical models, and computational schemes of various predictive models, speculate regarding their influence on individual sample- and population-based synthetic lethal interactions, discuss the pros and cons of existing SL data and models, and highlight potential research directions in SL discovery.

Knowledge-based system has become a useful tool to help clinicians in their diagnosis and decision-making during their routine practices. The objective of this present study was to develop a knowledge-based system applied to orthopedic pediatric disorders (rotational abnormalities, club-foot deformities and children with cerebral palsy). Based on patient data, two tree-based data mining approaches were studied to develop mathematical predictive models. The first one related to a deterministic standard decision tree model derived from one-dimensional morphological patient data. The second one deals with a belief decision tree model derived from high multi-dimensional motion capture data (kinetics (reaction forces) and muscle activities (EMG signals)). As results, deterministic predictive diagnostic models of rotational abnormalities and clubfoot deformities were presented and analyzed. On the other hand, a belief classification model of children with cerebral palsy was presented. We found that reaction forces and muscle activities of rectus femoris are main discriminators for recognizing diplegia and hemiplegia cases. More-over, the belief level of each classification result was also calculated and presented to take uncertain and imprecise characters of input data into account. Rule-based sets extracted from decision trees to develop diagnosis/classification algorithms were also presented. Validation of developed predictive models was performed on real patient cases. Web-based interfaces including step-by-step diagnosis, treatment and monitoring processes were also presented. To conclude, tree-based data mining approaches demonstrated their flexibilities and performances in developing deterministic and belief predictive models. Furthermore, they showed strong generalization ability to interpret the classification results. In fact, clinicians could use our knowledge-based system to take appropriate medical decisions in an efficient and convivial manner.

Medical data clustering is popular scientific approach for finding hidden patterns from large medical dataset. Medical experts utilised these patterns to make clinical diagnosis for likelihood of a disease. Clustering groups the data objects of dataset into different groups based on data similarity within group is higher than other groups. In this work, a hybrid PSO-GA algorithm is developed for medical data clustering based on 'particle swarm optimisation (PSO) and genetic algorithm (GA)'. Hybrid PSO-GA performance has examined against K-means, PSO and GA with respect to six popular medical datasets namely iris, thyroid, breast cancer, heart, diabetes and pima adopted from UCI machine learning repository over three criterion namely, i.e., sum of intra cluster distance, error rate and CPU running time. Tabular and graphical results of simulation confirm hybrid PSO-GA technique for medical data clustering is superior against K-means, PSO and GA.

The rapid evolution of machine learning (ML) and big data analytics has modernized medical decision-making procedure, offering promising path for improving classification precision and ultimately, patient outcomes. This research inspects methodologies for enhancing the classification accuracy of medical decision systems by leveraging ML algorithms and big data analytics procedure. In this study, a broad evaluation of existing literature on ML applications in healthcare and medical decision-making is carried out to discover current challenges and potential areas for improvement. The research explores the integration of diverse data sources, including electronic health records (EHRs), medical imaging, genomic data, and patient-generated data, to build robust predictive models. Moreover, the research emphasizes the importance of interpretability and transparency in ML models for medical decision-making, particularly in critical healthcare settings where the rationale behind predictions is crucial. Techniques for model explainability, such as feature importance analysis and model-agnostic interpretability methods, are explored to enhance trust and adoption of ML-driven decision systems by healthcare professionals. Furthermore, the study investigates advanced ML algorithms such as deep learning, ensemble methods, and feature engineering techniques to extract meaningful patterns from large and complex medical datasets. Through experimentation with real-world medical datasets, the efficacy of these algorithms in improving classification accuracy is evaluated and compared against traditional methods. The result of this research contributes to the advancement of ML-driven medical decision systems by providing insights into strategies for improving classification accuracy, thereby facilitating more exact diagnosis, prognosis, and treatment recommendations. Ultimately, the integration of ML and big data analytics holds immense potential for revolutionizing healthcare delivery and improving patient outcomes.

Molecular biology experiments are generating an unprecedented amount of information from a variety of different experimental modalities. DNA sequencing machines, proteomics mass cytometry and microscopes generate huge amounts of data every day. Not only is the data large, but it is also multidimensional. Understanding trends and getting actionable insights from these data requires techniques that allow comprehension at a high level but also insight into what underlies these trends. Lots of small errors or poor summarization can lead to false results and reproducibility issues in large data sets. Hence it is essential we do not cherry-pick results to suit a hypothesis but instead examine all data and publish accurate insights in a data-driven way. This article will give an overview of some of the problems faced by the researcher in understanding epigenetic changes (which are related to changes in the physical structure of DNA) when presented with raw analysis results using visualization methods. We will also discuss the new challenges faced by using machine learning which can be helped by visualization.