Research Toolbox

RStan is AIMMLab's primary tool for full Bayesian statistical inference. Built on the Stan probabilistic programming language, it implements Markov Chain Monte Carlo (MCMC) sampling specifically the No-U-Turn Sampler (NUTS) alongside variational Bayes approximation and penalised maximum likelihood optimisation. In AIMMLab, RStan is used extensively for calibrating compartmental ODE models (SVEIR, SEIR, and their variants) to infectious disease surveillance data, estimating transmission parameters with quantified uncertainty, and validating epidemic models across diseases including influenza, COVID-19, mpox, cholera, and avian influenza (H5N1).

EndNote is AIMMLab's reference management platform for systematic reviews and large-scale evidence synthesis. It supports import from major databases: PubMed, Scopus, Web of Science, EMBASE and provides built-in deduplication, title/abstract screening workflows, and seamless integration with word processors for citation and bibliography management. In AIMMLab's systematic review of behavioural-epidemic modelling studies (N=376 included studies from an initial corpus of 10,000+), EndNote was central to the de-duplication and screening pipeline. It is also used across lab manuscripts submitted to high-impact journals in mathematical biology, public health, and epidemiology.

Natural Language Processing (NLP) and sentiment analysis are central to AIMMLab's work on the social dimensions of epidemics from tracking public opinion on vaccines and treatments to detecting misinformation and measuring stigmatisation of marginalised communities. Social media platforms have been successfully applied across behaviour analysis, spam detection, electoral prediction, event detection, and public health monitoring.

AIMMLab uses two complementary approaches. Sentiment analysis identifies the emotional tone of text classifying content as positive, negative, or neutral and assigning intensity scores to gauge public opinion, track risk perception, and monitor community responses during health crises. Topic modelling (using algorithms such as Latent Dirichlet Allocation and BERTopic) uncovers the underlying themes in large document corpora without prior labels, revealing what people are discussing, not just how they feel. Together, these methods provide a powerful lens: topic modelling reveals major areas of discourse, while sentiment analysis assesses the emotional stance within each theme.

AIMMLab has applied transformer-based models (including BERT, RoBERTa, and domain-adapted variants) to Twitter/X data for gender recognition, COVID-19 ivermectin discourse analysis, mpox stigmatisation tracking, and cross-country comparative studies of vaccine hesitancy across 15 countries.

Systematic and scoping reviews are how AIMMLab members turn a broad, scattered literature into a structured evidence base often as the first step of a PhD or MSc project. A scoping review maps a broad or emerging field (what has been studied, what methods and populations are common, where the gaps are), while a systematic review answers a focused question (does one method outperform another) and typically includes formal quality/risk-of-bias appraisal. AIMMLab reviews follow the PRISMA-ScR reporting guideline and a five-stage workflow: frame the question with the PCC framework (Population, Concept, Context), build and document a reproducible multi-database search strategy, screen records in stages with full audit trail (identification → screening → eligibility → included), extract data into a structured charting matrix (one study = one row, one variable = one column), and synthesise the results into an evidence map highlighting gaps and cross-cutting themes.

In AIMMLab's scoping review of AI applications in healthcare quality and patient safety, this workflow took an initial search of 43,394 records across five databases (PubMed, Scopus, Web of Science, IEEE Xplore, CINAHL) down to 275 included studies, mapped across four application domains (diagnostic AI, predictive analytics, clinical decision support, economic/value assessment) and synthesised into five cross-domain barriers to clinical translation.