A recent review, published in Mathematical Biosciences and Engineering, has explored the current standing of computational modelling and machine learning approaches for cancer research and drug discovery.
Cancer and AI
Cancer is an extremely dynamic and complex disease. It is one of the leading causes of death worldwide and poses a significant challenge for healthcare systems. The demand for effective therapies, particularly personalised therapies, has never been more apparent. Unfortunately, the majority of novel cancer therapeutics consist of failed drug trials, with most cohorts refractory to current treatment.
Advances in current technology have provided us with the opportunity to efficiently process extensive datasets in order to streamline drug development processes. Machine learning (ML), a subfield of artificial intelligence (AI), can predict future outcomes through learning from unstructured or structured data, identification and classification or hidden patterns. This tool has become of interest particularly to the pharma industry to improve aspects of the drug discovery pipeline. The availability of large ‘omic’ datasets using AI models can be rapidly screened to identify novel drug targets or testing response, providing an opportunity for implementation of AI throughout the cancer diagnosis and treatment process.
Modelling facilitates our understanding of an issue by incorporating essential aspects, functions and interactions of a system. The resolution of the model must be appropriately matched to the research question that is driving the investigation. The level of detail is important to ensure that the question can be sufficiently answered. For researchers to design computational models to investigate cancer pathophysiology, there first requires a deep understanding of the interplay between single cells and the tumour microenvironment. The authors also note that combining different methods often strengthens the biological relevance of the model. Computational models allow deep and thoughtful exploration of certain paradigms that cannot be tested on laboratory animals or humans. Types of models include:
- Agent-based models – commonly employed in cancer studies, each cell can be modelled as an independent autonomous entity
- Lattice-based models – spatially restrict cells to a grid network, allowing evaluation of their spatial resolution
- Cellular automaton (CA) models – form of lattice-based modelling, often employed to replicate monolayers or multi-cell solid tumour structures
- Lattice-gas CA models – form of lattice-based modelling, for demonstrating invasive capacity of tumour cells
- Cellular Potts models – forms of lattice-based modelling, each cell is represented by multiple lattice sites
- Centre-based off-lattice models – represent cells as points or spheres
The team express the importance of properly training models to avoid bias and ensure algorithms are cross validated. Proper model development is vital in establishing sensitivity with relevant data and populations.
Importantly, computational simulations, with the combination of extensive datasets, have also improved our understanding of resistance mechanisms. Moreover, high-throughput approaches have also enabled the incorporation of the pharmacokinetics and pharmacodynamics of drug treatment. The ultimate aim of implementing these modelling techniques is to facilitate more effective drug design by overcoming intricacies associated with cancer biology.
Drug discovery and development requires lengthy processes to identify therapeutically targetable elements of disease and depends on extensively high-quality datasets in order to produce a safe and efficacious drug. Machine learning not only provides a high-throughput approach for data analysis and storage, it also increases the likelihood of developing a successful product.
Predicting the mechanism of action of new drugs using neural network automation and evolutionary algorithms could have a profound impact on the pharma industry. In addition, experts have developed various ML algorithms for drug discovery. These include supervised learning algorithms of support vector machine (SVM) and random forest (RF). Researchers can specifically use these approaches for ligand or structure-based virtual screening.
Experts can use AI in the following drug discovery and development processes:
- Prognostic and pre-clinical development – biomarker prediction and high-throughput screening of cancer responses to novel therapeutics
- Target identification, prioritisation and validation – to identify disease targets and associated genetic markers
- Compound screening – investigators can use experimental and virtual methods to screen extensive compound libraries to determine the efficacy
- Clinical trials – to identify appropriate patient populations
In order to progress our understanding of cancer biology and develop new treatments, AI provides a sophisticated tool for processing biomedical literature and mining extensive datasets. Many pharma companies are already investing in these approaches in order to fast-track drug discovery. The significance of AI is apparent in all aspects of cancer prognosis, diagnosis and drug development. However, the authors believe that the next main question to be answered is whether AI-based methods can conceive the heterogeneity of cancer and combine multiple types of data to find their biological relations and enhance the prediction of effective drugs.
Image credit: By freepik – www.freepik.com