Impact of DNA on diagnostics
Pharmacogenomics (PGx) is a conglomeration of pharmacology and genomics for personalized treatment. It considers drug-drug, drug-gene, and gene-gene interactions. In principle, gene expression controls the overall influence of drugs. With artificial intelligence (AI), predictive PGx will provide information ahead of time.
AI-powered algorithms can be used to test any life-saving drug, targeting the community-level impact of PGx implication, even on a population. A drug’s interaction with the human body depends on how the medication was taken and which part of the body was targeted. Once the drug is consumed, the body responds by targeting it to the site of action. AI includes a fundamental paradigm called machine learning (ML), which engages a field of deep learning (DL), creating artificial neural networks (ANN). The DNA can impact multiple steps in the PGx process that influence drug response, and different types of ANNs can play a significant role in each step, as given below:
Drug receptors – The DNA determines the type and density of receptors, thereby impacting drug response and dosage. For example, treating breast cancer patients with a monoclonal antibody—like Trastuzumab emtansine (T-DM1)—demonstrate their dependence on the human epidermal receptor 2 (HER2) expression. Overexpression of HER2 is a well-identified prognostic and predictive biomarker in cancer biology. The drug, T-DM1, targets HER2 receptors on cancerous cells inducing apoptosis and mitotic catastrophe. However, in conditions where the tumour cells do not have a sufficient expression of HER2, the effectiveness of T-DM1 is affected (Figure 1). AI can measure the binding affinity of a drug through drug-target binding affinity (DTBA) using web applications and the similarity ensemble approach (SEA). Many well-known strategies involve ML and DL (such as KRONRLS, SimBoost, and others).
Drug uptake – Regulations over gene expressions govern the measure of drug uptake. For example, statins are a type of drug that acts in the liver to lower cholesterol levels. Statins are taken to the liver by a transporter protein called organic anion transporter 1B1 (OATP1B1 or SLCO1B1). A decrease in gene expression might result in reduced statin uptake. In high doses, simvastatin might accumulate in the blood, causing muscle weakness and pain. That is why before prescribing simvastatin, it is advisable to conduct a genetic testing for the SLCO1B1 gene (Figure 2). Algorithms for drug design include molecular descriptors (such as SMILES strings), potential energy measurements, and coordinates of atoms in 3D to generate feasible molecules via deep neural networks (DNN) and predict their properties.
Drug breakdown – Gene involved in metabolizing the drug regulates the manner and rate of drug breakdown. For example, the study of amitriptyline—an anti-depressant drug—is influenced by a couple of genes called cytochrome P450 family 2 subfamily D member 6 (CYP2D6) and cytochrome P450 family 2 subfamily C member 19 (CYP2C19). The genetic behaviour of these two genes should determine the dosage of the drug, as it might differ from person to person (Figure 3). ANN-based models, graph kernels, and kernel-ridge-based models were used to predict the acid dissociation constants of compounds for efficient breakdown.
Targeted drug development – Drugs require structurally, functionally, and stoichiometrically active proteins for the intended action. Genetic mutations can result in the creation of non-functional proteins, rendering the drug ineffective. One must consider such drug-gene interactions to evaluate drug effects. For example, mutations in the cystic fibrosis transmembrane conductance regulator (CFTR) gene, impact the CFTR protein, thereby causing cystic fibrosis. The CFTR protein acts as a channel to move particles across the cells in the body, obstruction of which results in a disorder known as cystic fibrosis. The Ivacaftor drug forcefully opens this channel, thereby treating cystic fibrosis. However, mutation on the CFTR gene results in the absence of the cross-membrane protein. In this case, Ivacaftor will not have any impact on the patient (Figure 4). An integrated approach of model expert systems (MES) and ANN helps in the development of direct filling hard gelatin capsules of piroxicam per the specifications of its dissolution profile.
In all the above examples, we observed that genes significantly impact treatment strategy and dosage, especially in diseases that are neurodegenerative and those like cancer, along with others.
The next generation of biomedical science
The crucial themes shaping the next generation of biomedical sciences include pharmacogenomics, big data, analytics, and AI. Our focus is on the potential of AI to develop drugs, predict their efficacy, and introduce medical devices and treatment policies. We apply AI to create actionable insights from the patterns identified. These applications include algorithms using ML, deep learning constructs, and related technologies.
The paradigm shifts in health interventions
Over the years, there has been a paradigm shift in drug development. Traditional drug development followed a sequential pre-clinical discovery process, accompanied by clinical testing and implementation at a population level (Figure 5).
The personalization in drug development is more sequential, involving PGx assessment, choice, and testing. Extensive clinical data mining is based on genotype-enabled intervention for an individual (Figure 6).
The aim is to streamline diagnoses, treatment, and follow-up monitoring into one smooth process with coordinated transitions. In certain immunotherapeutic-oriented cell replacement therapies for cancer, a patient’s tumor is profiled for the existence of ‘neo-antigens’—mutations that attract the host’s immune systems to attack the cells harboring those mutations. Once the neo-antigens are found—either through allogenic transplantation using donor cells or autologous transplantation using the patient’s cells—they are harvested and sensitized to recognize the neo-antigens. When introduced into the patient’s body, these cells attract the host’s immune system to act on the cancerous cells containing the neo-antigen. The production of treatments in real-time based on the patient’s unique needs is termed as the ‘magistral’ production of treatments, as opposed to the traditional or ‘officinal’ production of treatment.
Impact of AI in magistral production on healthcare providers
AI is crucial in advancing personalized treatment production for various life-threatening diseases. Some of the ways to implement AI are:
AI-powered robotics - Efficient and precise manufacture of appropriate treatments, electrospinning has been widely leveraged in biomedical engineering to create tissue engineering scaffolds. Robot-aided electrospinning improves the control of quality parameters, like the diameter of nanofibers, and the rate of production, among others.
3D printing - The potential of manufacturing precise and relevant interventions with 3D printing is immense. The first US FDA-approved 3D-printed drug was made in 2015. A controlled drug release behavior can be based on genetic testing and designed by smart drug delivery systems (DDSs) that enhance the drug’s effectiveness for a particular patient. Various strategies have been used to accurately produce 3D structures, like ink-jet deposition, extrusion-based 3D printing, photo-polymerization, and the like.
N-of-1 trials - Once the treatment is designed, the immediate conduct of N-of-1 trials aided by AI-based pattern discovery will assess its impact on the patient and provide sophisticated treatment outcomes.
AI-based simulation studies - This explores, pre-empts, and anticipates possibilities that a treatment strategy might take.
Impact of PGx testing in treatment coverage on payers
Drug development is a highly regulated process with specificity, potency, and efficacy. One of the first successful examples of PGx testing is HER2 or NEU testing for Herceptin in the year 2005; since then, there has been good progress in PGx testing in the field of genetic testing for somatic mutations in lung and breast cancer.
A few areas where AI can augment PGx testing are as follows:
The AI-based economic framework - It focuses on the cost and outcome effectiveness of PGx testing. This would fuel ‘value-based purchasing’ for members and effective decision-making for payers.
Data-based outcomes – Large payer organizations in the US only cover the test if it comes back with a high-risk score, thereby creating uncertainties and unhappy members. However, data can augment PGx testing outcomes for an effective coverage policy by the payer. For instance, genomic health’s Oncotype DX® test helps oncologists segregate breast cancer patients most likely to benefit from adjuvant chemotherapy. If a PGx test result can be augmented with data-driven analysis of tumors of the same genomic profile, the scoring could be more accurate and reimbursement policies more satisfactory.
The contributions of AI in advancing PGx for treatment personalization in conditions like neurodegenerative disorder, cardiovascular diseases, and cancer, among others, will be profound. Not only will AI-based health products be the future, but the use of technologies like quantum computing and edge computing will be ever-increasing.