Overview

In recent years, the integration of artificial intelligence (AI) into the healthcare field has gathered significant momentum. AI has proven invaluable in automating repetitive and time-consuming tasks that are frequently involved in genomic studies, a field necessary to conquer in achieving personalised healthcare.

Many genetic diseases, including cancer, have complex etiologies that complicate the treatment process. Personalised healthcare plays a key role in overcoming this issue, as it accounts for the diverse genetic backgrounds of patients by customising medical interventions, resulting in improved treatment outcomes. 

AI has shown immense potential to accelerate the planning and implementation of personalised healthcare by enhancing diagnosis, streamlining drug development, and improving phenotype/genotype prediction.

Role of genomics in personalised healthcare

What is the human genome?

The human genome is the entire genetic sequence found in human cells required to code a human being.¹ A genetic sequence is equivalent to the DNA sequence, consists of over 3 billion nucleotides, that are organised into 23 pairs of structures known as chromosomes. 

Within this exhaustive amount of biological information, it is rather surprising to note that only 0.01% differs between individuals.²

Why is genomics important in personalised healthcare?

Although a 0.01% variation may seem minuscule and unimportant, this is what gives rise to the diversity among people, including, but not limited to, drug responsiveness, disease susceptibility, and physical performance. Due to such variations, standardised treatments often fail to achieve consistent effectiveness and outcomes among the general population. Thus, these genetic variations are key targets for achieving personalised healthcare.

Introduction to AI in healthcare

Key terminologies in AI

Before learning about AI in healthcare, it is important to understand how AI acquires its ability to mimic the human brain.

• Machine Learning (ML) is the process by which the machine, or AI, is programmed to learn to categorise datasets without a particular instruction. For example, the machine can be presented with two sets of genomic data, one from patients with Autism spectrum disorder (ASD) and the other without. The machine would then observe the data using hidden patterns to distinguish the two groups and apply these distinctions to real-life samples. ML can be done either supervised or unsupervised
• Deep Learning (DL) is a part of ML, where the machine observes the data without supervision. In the DL algorithm, AI finds patterns in the data, similar to how the neurons in the human brain work, using them as a metric to distinguish real-life datasets

Impact of AI on healthcare innovation

AI has impacted the field of healthcare through a range of different applications. Examples include:

• Computer vision – DL AI algorithms help to analyse cancer histopathological images to identify cancer cell types and prediction of oncogene mutation³
• Time series analysis – Although this is typically applied to temporal sequences, for example, electrocardiograms, it can also be used in nucleotide sequences to check for any abnormality or regulator function
• Natural language processing (NLP) – AI can be used to extract meaning from human language, which allows the prediction of rare genetic diseases related to language impairment
• Clinical trials – AI has helped clinical trials for newly produced drugs. It has been used in tasks that were previously time-consuming, such as trial design, protocol writing and filtering of patients

Applications of AI in genomics

Drug discovery and development

In recent years, the number of AI-discovered drugs has doubled since 2021 AI has driven this change by rapidly repurposing existing drugs or identifying new drug targets.⁴ Moreover, AI can screen vast amounts of molecular data to filter out target candidates that match the drug requirements, thereby accelerating and improving the accuracy of target validation.

The same applies to genomic data too, where DL AI algorithms have helped identify genetic targets in diseases such as Alzheimer’s.⁵ For example, Alphafold2, an AI program developed by DeepMind is used to predict protein structure from the gene identified, enhancing drug design and development.

Cancer treatment

In the treatment of cancer, AI can be combined with traditional screening methods or diagnostic approaches to improve treatment outcomes. For example, combining AI with biopsy examination has improved the detection and grading of prostate cancer, reducing clinicians’ workloads. In genomics, AI is used to understand the functions of both regulatory and protein-coding genes.

For protein-coding genes, DL-based meta-predictions predict the effects of genetic variants and thus guide their classification into cancerous or harmless variants.⁶ It is worth noting, though, that such AI usage is only limited to supporting the clinician’s decision-making process and prioritisation of patients who carry more significant variants, rather than leading the classification process by itself, as it is still inaccurate. One example of an AI programme that has spearheaded this change is PrimateAI.⁷

Variants for non-coding genes such as regulators and silencers, have far more complex effects than coding genes, which complicates the applicability of AI within a clinical setup. To predict enhancer role in cancer development, an AI system known as Explainable AI or XAI has been incorporated to provide greater transparency and accessibility to the generally opaque AI algorithms. This has greatly improved the performance of AI in healthcare, allowing personalized medication that specifically targets the gene regulator in question.²

Genetic disorders

Computer vision-based AI algorithms can detect facial abnormalities and predict the underlying genetic mutation. The process of matching phenotypes with underlying genetic mutations, known as genotype-phenotype matching, is essential for accurately diagnosing developmental disorders, as many of these disorders typically share similar phenotypic facial structures.

Studies have shown that an AI program called DeepGestalt significantly outperformed a human dysmorphologist in mapping clinical diagnosis with a molecular diagnosis.⁸ AI-driven genotype-phenotype matching can also detect small facial abnormalities that would otherwise go unnoticed by a human clinician, allowing a more diverse group of phenotypes to be considered in healthcare.

Benefits of AI-driven genomics in personalised healthcare

Improving accuracy in diagnosis

Other than improving genotype prediction, AI can learn to remove bias when identifying variants from genomic datasets (a process known as variant calling). Such biases are caused by sample preparation, technology, and biological influences, which were previously removed using more demanding statistical approaches.⁹

Improving treatment plans

Genotype-phenotype prediction is another approach attempted by AI, wherein algorithms predict complex human traits from patient’s genomic dataset alongside non-genetic information, including lifestyle, age, environment, etc.⁶ Such factors are considered in a risk model which helps to strategise future diagnosis and treatment plans. 

Challenges and ethical considerations

Data privacy and security

Increased dependency on AI raises vulnerability to cyberattacks, such as hacking, leading to greater risk in data privacy. Some bioinformatic companies may lack protective measures to prevent information leaks or are involved in selling patient information to other healthcare companies.¹⁰ Such issues are more likely to occur due to the insufficient regulations that protect healthcare data in the current industry landscape.

Fairness and transparency

Regardless of AI usage, transparency and explainability of healthcare have always been important ethical considerations in healthcare. With that said, the issue seems to be aggravated by AI usage, as studies show that patients can be highly concerned about the transparency of guidelines and regulations related to AI in healthcare.¹¹

It is important for clinicians, as well as healthcare companies, to inform the extent and nature of AI involvement in healthcare to remove such concerns, and respect patients’ rights to understand the treatment procedure and refuse treatment if they feel uncomfortable.¹⁰

Bias in data collection

Bias is another key issue in AI. Data collection during AI development is a source of biases, as data may be more focused on a particular ethnic group or gender. Appreciating underrepresented minority groups in AI design is crucial to ensuring model fairness and accurate performance in healthcare.

Summary

AI programs are a vital tool in rapidly processing vast amounts of genomic data to achieve personalised healthcare in many genetic diseases and disorders. By improving f diagnostic accuracy, accelerating drug production, and clinical trials, companies can streamline healthcare production that accounts for greater individuality. AI can also be used to predict phenotypes or protein structures from genetic variants, to further support strategised treatment plans. 

It is necessary, however, for such companies and clinics to consider the ethical complications such as data privacy and transparency. To achieve this, greater clarity and strictness in healthcare regulation is important, as in the future years, AI involvement in healthcare will continue to be more common.