Increased life expectancy in the developed world in the 20^th century can be attributed to several factors – sanitation infrastructure and clean water, increased food security, population-based healthcare systems, mass vaccination programs for the most prevalent infectious diseases of childhood and the use of antibiotics. The discovery of penicillin in 1928 by Alexander Fleming and its purification by Howard Florey and Ernst Chain in 1940 ushered in the ‘antibiotic age’ and laid the foundation for exploring the potential of a plethora of novel antimicrobials (Hutchings et al., 2019; Lima et al., 2020; Vila et al., 2020). Currently, it takes about 12 years to get a drug to the market for public use and the process is hugely expensive, with “median development costs for a new antibiotic exceeding $1 billion and sponsor costs of ~$350m to complete post-approval work and manufacture of the compound during its first 10 years on the market” (Wouters et al., 2020). However, not long after the first clinical use of penicillin, micro-organisms were observed to acquire antibiotic resistance through several different mechanisms (Christaki et al., 2020; Huemer et al., 2020; Larsen et al., 2022). The World Health Organization (WHO) has highlighted the priorities for tackling drug resistance in bacteria, viruses, parasites, and fungi, which requires a globally coordinated multi-sector approach (Tacconelli et al., 2018). Antimicrobial resistance has undeniably been aggravated by i) the extensive use of broad-spectrum antibiotics in agriculture, veterinary and medical practices, ii) in self-medication, iii) with poor diagnostics, and iv) with a heavy patient load on healthcare systems (Sulis et al., 2022). Antibiotic resistance is also spreading rapidly in the environment due to contamination of water bodies through the discharge of untreated domestic sewage, hospital wastewater, effluents from the antibiotic manufacturing units, open defecation, and mass bathing. These contaminated water bodies may lead to the massive growth of resistant bacteria and act as a pool for antibiotic resistance transfer (Davies and Davies, 2010; Perry et al., 2016; Hutchings et al., 2019; Christaki et al., 2020). Prevailing antibiotic resistance has led to a rapidly diminishing pool of bioactive compounds and plausible solutions to combat the challenges posed by pathogen resistance include i) the modification of existing antibiotic classes (repurposing), ii) exploring new structural classes, and iii) the development of prophylactic vaccines. Recent antimicrobial development regimes use resistance gene(s) screening, whole genome(s) sequencing, and correct pathogen detection approaches (Vila et al., 2020; Wang et al., 2022).

Increased antibiotic resistance in micro-organisms, specifically amongst bacteria, has stimulated the development of novel antimicrobials and vaccines using Artificial Intelligence (AI), Machine Learning (ML) and Neural Networks (NN) (Dalsass et al., 2019; Stokes et al., 2020; David et al., 2021; Sahayasheela et al., 2022). The AI-based learning techniques can be categorized into three classes: 1) supervised learning, where the prediction model learns from the previous examples of classification, 2) unsupervised learning, where the prediction model learns by exploring different patterns in the training data, and 3) reinforcement learning, where the prediction model learns by implementing a scheme of operations depending upon reward sequences and penalty. In general, development of an AI-based prediction model requires data pre-processing followed by encoding, feature extraction, training/learning from the features, and testing, validation and evaluation on unseen data with the help of different AI platforms available to the developer. Some of the widely used AI platforms include TensorFlow, Google AI, Microsoft Azure, OpenAI, amongst many. TensorFlow is an open-source ML platform that offers a broad range of AI algorithms to build and design deep learning models from training data of interest. Google AI is a division of Google, focusing on artificial intelligence through the Google Cloud AI platform that hosts several AI tools armored through generative AI (for prototyping and testing generative AI models), vertex AI (to create, train, test, monitor, tune, and deploy ML and AI models), natural language AI (derive insights from unstructured text using Google machine learning), translation AI (multilingual with fast, dynamic machine translation), vision and video AI (detect objects, understand text, videos and more), and so on. The Google AI platform is another extensive suite of open-source AI algorithms with a fast and easy implementation to build used defined prediction models. Microsoft Azure is a comprehensive set of cloud services that offers solutions for AI analytics through its AI tools and frameworks, equipped with multi-layered data protection. OpenAI is an AI research and deployment driven company focusing on benefiting humanity through artificial general intelligence, controlled through a unique capped-profit model, and it delivers very powerful models (e.g. Generative Pre-trained Transformer 4, GPT-4) that are trained to generate ‘human-like’ text and to enable multi-domain applications. ChatGPT and DALL-E2 are among the famous models developed by OpenAI. OpenAI is a potential game-changing state-of-the-art AI platform with the potential to translate languages, write articles and answer operator questions. An AI platform can enable users to develop their own deep leaning models and to utilize pre-trained models for their specific research problems. Trained prediction models can be evaluated with different evaluation matrices that include mean absolute error (MSE), root mean square error (RMSE), various statistics derived from confusion matrix, accuracy, precision, area under curve (AUC), receiver operating (ROC) curve, etc. The prediction model can be implemented for practical purposes after the desired accuracy - as defined by the different evaluation matrices - is achieved (Wang et al., 2019; Yang et al., 2019; Jiang et al., 2020).

The use of AI computational approaches validated fast-tracked novel drugs, drug repurposing and vaccine designs during the COVID-19 pandemic, along with developing infection diagnostics (Shrock et al., 2020; DeGrace et al., 2022; Thomas et al., 2022). AI assisted prediction of B-cell and T-cell epitopes and novel vaccine candidates is rigorously driving immuno-informatics-based approaches towards developing refined AI- and ML-based prediction servers. AI systems reason to find the microbial components that are unlikely to mutate or alter, to guarantee that a vaccine remains effective for a long duration. Computational analysis aided with ML algorithms have played a pivotal role in vaccine development. For example, AI-based approaches can provide structural and molecular insights on SARS-CoV2 viruses and predict the viral components that can trigger potentially protective immune responses and interpret experimental findings. By combining data from multiple experimental and real-world sources, AI can monitor the genetic changes (mutations) in the SARS-CoV2 genome over time to maximize future vaccine efficacy (Waltz, 2020). Different AI and ML techniques have already demonstrated their potential in diverse healthcare and biomedicine related fields, for example by accelerating the discovery of novel antimicrobials in a cost-effective manner, reducing expenses on equipment, synthesis, and human resources. Some well-known AI/ML techniques include support vector machine (SVM), logistic regression (LR), random forest (RF), and different types of neural networks, including multi-layer perceptron (MLP), recurrent neural networks (RNN), convolutional neural networks (CNN) and deep neural networks (DNN), amongst others (Wang et al., 2019; Jiang et al., 2020; Stokes et al., 2020; Vila et al., 2020; Sahayasheela et al., 2022; Thomas et al., 2022). The SVM algorithms implement a supervised statistical leaning approach for linear and non-linear data classification and regression analysis. The LR algorithms utilize standard logistic (sigmoid), and probability (log-odds) functions derived from the explanatory variables to predict the outcome variables. The RF algorithms implement an extensive multiple decision tree approach to learn from the training data and perform the prediction. These algorithms utilize bagging and feature randomness to generate a random forest of decision trees. The MLP algorithms implement a completely connected feed-forward artificial neural network with at least three hidden layers along with an input layer and an output layer. The flow of information passes from input layer to output layers through hidden layers, controlled by an activation function. The RNN algorithms are the advanced version of the MLP algorithms where the flow of information can also occur in a cyclical manner in the hidden layer to impart recurrent learning from the input layer. Likewise, the CNN algorithms are regularized MLP algorithms. The regularization nullifies the possibility of over fitting of data during training due to fully connected layers in the MLP. The CNN algorithms implement convolutions functions instead of matrix multiplication functions. The DNN algorithms broadly cover MLP, RNN, CNN, and several other neural network algorithms. The neural network that implements multiple hidden layers can be categorized as DNN. It should be noted that the list of different AI techniques mentioned above is not exhaustive and providing their fundamental details are beyond the scope of this mini-review.

It is worth mentioning also that AI and ML are intricately related (as both are supervised learning), but conceptually different. AI is a broad field of computer science focused on creating machines or systems that can perform tasks requiring human intelligence, such as problem-solving, reasoning, and understanding natural language. ML is a subset of AI that focuses on the development (training, fine tuning, testing, and deployment) of task-orientated prediction models. ML involves training machines to learn from data and make predictions or decisions without explicit programming. In essence, ML is a technique used within the broader field of AI to achieve ‘intelligent behaviour’. Hereon in, we will use the broader term AI to embrace the diverse subfields of AI and avoid any ambiguity and inconsistency. The field of AI is anticipated to flourish in coming years to become a vital tool to combat microbial infectious diseases. Considering the principles of AI based techniques, the focus of this mini-review is to provide an overview of current AI practices in the discovery and development of vaccines.

The 21^st century has seen unparalleled and innovative growth and development of cutting-edge technologies to benefit society (Jaffee et al., 2017; Vetrano et al., 2022). AI and ML technologies have made significant contributions to the healthcare and medicine sectors, with the market expected to expand to $45.2 billion by 2025 (Kruk et al., 2018; Ahmed et al., 2020). Recent advances in high throughput experimental techniques have generated a considerable amount of ‘big data’ in the healthcare and biomedical sectors. These data sets fuel AI to deliver highly accurate projections and predictions in the fields of vaccine development and drug discovery. Notably, ~40% of companies engaged in drug discovery are using various AI techniques to identify drug targets and for novel drug design (Lee et al., 2022; Zeng et al., 2022; Blanco-González et al., 2023).

The emergence of new infectious pathogens and increased antimicrobial resistance amongst existing pathogens has accelerated AI-assisted vaccine development in recent years (Thomas et al., 2022). The availability of high throughput genomic and proteomic data garnered from various infectious diseases could serve as a catalyst for developing reliable AI driven predictive models. Vaccine design benefits from an AI-assisted better understanding of the pathogen infection cycle at the genetic, molecular, and cellular levels (Ong and He, 2022; Goodswen et al., 2023).

Identification of potential antigen(s) is the important step for vaccine development. Conventional experimental approaches for antigen detection were laborious and time intensive and the arrival of Reverse Technology (RV) has revolutionized the field of vaccine development. RV is a computationally assisted genome-based approach of vaccine design that circumvents the necessity of developing bacterial cultures to prioritize vaccine targets. The analysis of pathogen genomes with computational bioinformatics has benefitted the screening of potential vaccine candidates significantly. Recent advances in RV techniques are assisted by the incorporation of modern AI techniques at various stages of vaccine development, and these AI based models are beginning to expedite the discovery and optimization of potential vaccine candidates. The contribution of AI computational tools to the development of SARS-CoV-2 vaccines has demonstrated the potential of AI for vaccine development against different microbial pathogens. One of the major challenges for AI is to accurately identify the potential antigens that can trigger host immune responses from amongst the thousands of pathogen components. In the following section we discuss the relevance of AI in vaccine development, its checkpoints, and potential solutions.

The WHO asserts that antimicrobial resistance (AMR) is one of the top 10 global threats to human health and development. In 2022, at least 1.27 million deaths per year are directly attributable to AMR (Murray et al., 2022) and by 2050, 10 million deaths are estimated per annum from infections caused by AMR bacteria, with economic healthcare-associated costs in the trillions of dollars (USD). Vaccines represent a prophylactic strategy to combat AMR, although it is worth mentioning that vaccine development against certain pathogens is hampered by a lack of understanding of the mechanisms of infection and of how immunity develops. The accelerated development of COVID-19 vaccines has shown how state-of-the-art biomedical technologies can serve society during pandemic crises. It is inevitable that new and developing AI and ML techniques will be used to develop vaccines for infections caused by many viruses, bacteria and parasites that have proved particularly intractable. Implementation of advanced AI techniques in research dates to the start of 21^st century. Undoubtedly, the development of AlphaFold (Higgins, 2021; Jumper et al., 2021; Binder et al., 2022; Varadi et al., 2022), a deep ML-based method for protein structure prediction, and the discovery of halicin, a potent antibiotic compound identified with a deep learning approach (Stokes et al., 2020) has imparted a belief amongst scientific communities of the potential of AI driven methods and technologies.

In summary, the integration of AI in vaccine development represents a transformative leap forward, expediting the process, enhancing precision, and broadening our understanding of infectious diseases. Whilst AI will significantly streamline the development of vaccines, we emphasize that these technologies are not a panacea. The integration of AI should always complement and not replace, rigorous ‘wet’ laboratory experimentation and animal and human trials, which remain indispensable steps in ensuring vaccine safety, efficacy, and are essential for regulatory approval. Nevertheless, as we continue to harness the power of AI-driven tools, the future holds great promise for the rapid creation of effective vaccines, particularly for emerging biological threats. We predict that AI assisted research and innovation in antimicrobial and vaccine development programs will ultimately improve global human health and increase life expectancy.

RahulK: Writing – original draft, Writing – review & editing. RK: Writing – review & editing. MC: Writing – review & editing.