Image-based profiling is a maturing strategy to extract information present in biological images. We summarise a recent article, published in Nature Reviews Drug Discovery, which explored the use of machine learning strategies to better leverage image-based information.
In drug discovery, the evaluation of the efficacy and safety of all candidate compounds in humans is ethically and practically not feasible. Therefore, researchers have to use simpler model systems, e.g., cells to map clinical efficacy and safety. Screening assays are often used to test thousands to millions of small molecules to identify target hits.
Profiling is an alternative strategy to screening. This approach aims to capture a wide variety of features, which may have relevance to a disease or potential treatment. Profiling represents model systems with a more comprehensive set of features. Out of the high-dimensional profiling techniques available, image-based profiling using automated microscopy is the least expensive. It also offers single-cell resolution and can capture important heterogenous cell behaviours. Profiling in the early stages of drug discovery can reveal key biological readouts that can be used for subsequent phases of screening.
Applications of image-based profiling
Image-based profiling requires images of biological samples that represent different cases or treatment conditions. The images are processed to extract features, which are then aggregated into profiles. Below are just some of the applications of image-based profiling in the drug discovery process:
- Profile-based phenotype discovery and screening – Identifying a disease-associated phenotype in images. Several strategies exist for identifying a cellular disease state with a profile that differs from that of the healthy state. The drug industry as a whole has begun to adopt image-based profiling to inform target identification and validation, phenotype discovery and assay development before screening.
- Lead generation – Hundreds of hits are then narrowed down to just a few candidates. The various applications of image-based profiling for lead generation can use either unbiased assays or customised assays with relevant biomarkers. This step involves hit expansion, lead optimisation and predicting assay activity and toxicity.
- Identifying the mechanism of action (MOA) – Elucidating the MOA of a drug provides a deeper understanding of its biological activity and increases its chances of clinical approval. There are three broad categories of image-based profiling approaches to determining the MOA. Specifically, guilt-by-association with annotated compounds, guilt-by-association with perturbed genes and a rescue experiment.
Driven by the recent availability of increasing volumes of images, the field has begun to turn to machine learning, and specifically deep learning, to improve extraction of relevant signals from profiles. The promise of improved profile resolution through rich data and machine learning has rekindled industrial interest in phenotypic profiling. However, this comes at a high cost, in terms of recruiting and retaining experts. Machine-learning strategies also risk problems such as overfitting, technical artefacts or confounders in the data, and bias.
The authors expect that advancements in the field of image-based profiling (both computational and biological) will rapidly progress in the next five years. In terms of computation, image-based profiling will be among the major beneficiaries of advancements in deep learning algorithms that are already beginning to accelerate drug discovery. On the biology side, scientists are adopting more complex model systems for their image-based profiling, such as organoids. Importantly, with increasingly available high-quality image datasets, more powerful machine-learning methods will emerge which will drive the cycle of discovery.