Artificial intelligence for omics data analysis

Abstract

Recent technological advancements have vastly improved access to high-throughput biological instrumentation, sparking an unparalleled surge in omics data generation. The implementation of artificial intelligence techniques is revolutionizing omics data interpretation. The BMC Methods Collection "Artificial intelligence for omics data analysis" will feature novel artificial intelligence approaches leveraging multi-omics data to accelerate discoveries in personalized medicine, disease diagnostics, drug development, and biological pathway elucidation.

Main

In recent years, technological advancements have significantly boosted the accessibility of high-throughput biological instrumentation for researchers. This surge has led to an unprecedented rate of biological data generation, marking the dawn of the big data era [1]. Driven by the aspiration for a comprehensive understanding of biological systems, researchers now routinely conduct omics studies, encompassing genomics, transcriptomics, epigenomics, proteomics, and metabolomics, which generate vast amounts of data that hold crucial information about biological processes and disease mechanisms. However, single omics data alone may sometimes struggle to fully elucidate the complexities of biological phenomena [2]. Therefore, integrating data from multiple omics sources can offer a more comprehensive understanding of biological systems by capturing interactions between different molecular layers. As a result, multi-omics approaches are gaining popularity due to their potential to provide a more holistic view of biological mechanisms or diseases by extracting, analyzing, and interpreting hidden information that single technologies cannot reveal [3].

Artificial intelligence approaches in omics analysis

Traditional statistical modeling has long been the default choice for analyzing and interpreting big data. However, in recent years, artificial intelligence (AI) technology has gained popularity across various fields [1]. This surge in popularity can be attributed to the evolution of data types from traditional structured data to non-structured, semi-structured, and heterogeneous architectures with diverse characteristics. Furthermore, the demand for novel insights into biological mechanisms has raised the standards and requirements for the depth and accuracy of omics analysis.

AI was formally defined at the Dartmouth conference in 1956 [4]. After that, it developed rapidly and it now encompasses a range of techniques, including machine learning (ML) and deep learning (DL), that enable computers to learn from data and make predictions or decisions. Specifically, ML focuses on developing algorithms and statistical models that enable computers to perform tasks without explicit programming. Algorithm selection is therefore pivotal, and they can be categorized into supervised, semi-supervised, and unsupervised [5]. DL, a subset of ML, employs neural networks composed of hidden layers that perform various operations to uncover intricate representations of the data. This approach has significantly improved the performance of classifiers, surpassing that of traditional ML algorithms, particularly in scenarios involving large-scale datasets with high dimensionality [5].

The implementation of AI techniques has certainly revolutionized the way researchers derive insights from omics data. For example, the recent developed genomic language model (gLM), trained on millions of metagenomic scaffolds to learn the latent functional and regulatory relationships between genes, has proven to be a potent and promising method to close the gap between genomic-context and gene sequence-structure–function [6]. In another recent study, MethylBoostER (Methylation and XGBoost for Evaluation of Renal tumors), a ML model based on the XGBoost (eXtreme Gradient Boosting) library, has been effective in differentiating pathological subtypes of renal tumors, using DNA methylation markers identified in large tissue datasets [7].

As previously mentioned, the interpretation of single omics data often falls short in explaining complex biological phenomena comprehensively, making it challenging to meet the growing research expectations. However, by integrating multiple omics datasets, researchers can gain a more comprehensive understanding of biological systems. AI techniques have become instrumental in this regard, allowing researchers to manage the high dimensionality and heterogeneity of multi-omics data. This approach not only uncovers hidden patterns but also facilitates the prediction of biological outcomes, thereby accelerating biomedical research and paving the way for personalized medicine [8]. For example, the recently implemented Molecular Twin, a novel AI platform integrating multi-omics data, has proven to be effective in predicting outcomes for pancreatic adenocarcinoma patients [9].

Challenges and perspectives

The accumulation of a large amount of biomedical data and the integration of multi-omics through AI will inevitably bring huge benefits to research, eventually leading to personalized medicine. However, despite the progress made by AI in various biomedical realms, numerous challenges remain [1]. They include but are not limited to the management and integration of high volume and heterogeneous multi-omics data, the expertise required for implementing AI approaches and interpreting AI-driven insights, and the critical task of maintaining data quality and achieving reliable generalization. More details are provided below.

Heterogeneity, outliers and missing data imputation

Multi-omics data from different high-throughput sources are usually heterogeneous and noisy. Some omics are more prone to generate sparse data than others and some datasets lack a large number of values, which hinders the integration of multiple datasets [10, 11]. Data preprocessing steps, such as normalization, batch correction, missing value imputation, and outliers detection are crucial for ensuring the quality and reliability of omics data analysis results [1].

Interpretability and explainability

AI models, particularly DL models, are often regarded as "black boxes" due to their complex architectures and lack of interpretability [1]. A transparent and explainable AI algorithm is essential to its final clinical translation and application. On March 15, 2024 the Food and Drug Administration (FDA) published the "Artificial Intelligence and Medical Products: How CBER, CDER, CDRH, and OCP are Working Together," which represents the FDA's coordinated approach to AI. This paper is intended to complement the "AI/ML Software as a Medical Device Action Plan" and represents a commitment between the FDA's Center for Biologics Evaluation and Research (CBER), the Center for Drug Evaluation and Research (CDER), and the Center for Devices and Radiological Health (CDRH), and the Office of Combination Products (OCP), to drive alignment and share learnings applicable to AI in medical products more broadly (https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-software-medical-device).

Overfitting and generalization

Overfitting, where a model performs well on the training data but fails to generalize to new, unseen data, is a common challenge in AI-driven omics data analysis. Techniques such as cross-validation, regularization, and ensemble learning are used to mitigate the risk of overfitting and improve the generalization performance of AI models [5].

Curse of dimensionality

Most multi-omics datasets suffer from the classical ‘curse of dimensionality’ problem, i.e. having much fewer observation samples than multi-omics features [11].

Computational and storage cost

The use of AI for multi-omics analysis comes with computational and data storage costs. Most algorithms require high computation power and large volumes of storage capacity to save the logs, results, and analysis [1].

Addressing these issues necessitates Findable, Accessible, Intelligent, and Reproducible (FAIR) solutions, designed for users with and without computational background [12]. These solutions should facilitate biomarker discovery and disease prediction with high precision by leveraging both existing and newly generated multi-omics data alongside demographic and clinical information, uncovering insights often overlooked by traditional statistical and bioinformatics methods. For example, the recent introduction of SLIDE (Significant Latent Factor Interaction Discovery and Exploration), an interpretable latent factor regression-based machine learning approach implemented for ubiquitous biological discovery from high-dimensional multi-omics datasets, overcame some of the previous challenges. While most current methods, such as black-box DL approaches or classification/regression techniques, focus primarily on prediction, preventing them from offering insights into actual mechanisms of complex molecular, cellular or organismal phenotype, SLIDE incorporated nonlinear relationships and came with rigorous guarantees regarding identifiability of the latent factors and corresponding inference [13].

Using AI to address research and clinical needs

However, a critical question remains: which AI approach or algorithm is most suitable to address a specific research question or clinical need? The choice of the appropriate AI approach profoundly influences outcome prediction accuracy, biomarker discovery, and stratification of patient heterogeneity. By applying suitable AI techniques, avenues can be opened for broader biomedical research, ultimately leading to personalized interventions and identification of novel treatment targets [3]. The widespread adoption of these advancements holds immense potential for enhancing public health initiatives worldwide.

Acknowledging the importance of this field, the BMC Methods Collection “Artificial intelligence for omics data analysis” (https://www.biomedcentral.com/collections/aioda), focuses on publishing innovative AI approaches using multi-omics data to accelerate discoveries in areas like personalized medicine, disease diagnostics, drug development, and biological pathway elucidation. We invite researchers to submit their work in these areas, contributing to the advancement of AI-driven omics data analysis and its applications in various fields of biological and medical research.