Scientific workflow for hypothesis testing in drug discovery: Part 1 of 3

In drug discovery and biological research, the scientist’s workflow often follows a structured and iterative approach to ensure accuracy, reproducibility and scientific integrity. This article outlines the various stages of such a workflow, from hypothesis generation to data cleaning and interpretation, referencing the project workflow diagram (Figure 1).

At the heart of any research is the scientific question. This question shapes the direction of the entire investigation and helps to focus the analysis on specific hypotheses. Following this, it is important to generate hypotheses based on prior research and available datasets. These datasets, which may be either public or proprietary, provide the foundation for all subsequent analysis. Public datasets, while larger, often require careful scrutiny as they may contain noise or inconsistencies that need to be accounted for before beginning analysis.

On the other hand, proprietary datasets tend to be smaller and are generated under more controlled conditions, such as those derived from in-house assays or experiments conducted within a lab. These proprietary datasets typically do not require as much validation as public data but still necessitate a solid understanding of their experimental design and data generation methods.

Figure 1: High-level workflow for early drug discovery

Once the raw data has been gathered, the next step is to gain a thorough understanding of the data. This includes verifying the experimental design and understanding how the data was generated –whether it was from RNA sequencing, mass spectrometry, or other biological assays. Public datasets, linked papers or supplementary documents often provide essential context about the experimental setup. These linked materials may describe the cell lines used, the specific conditions under which experiments were performed, and any potential limitations of the data. Without this information, the interpretation of the dataset may be speculative or lead to inaccurate conclusions.

The next critical step is the “sanity check” phase, where the integrity of the dataset is assessed. This process involves checking for biological inconsistencies or errors in the data, such as the presence of genes that should not be expressed in certain tissues. For example, while working on diabetes research, one can include proteins involved in bone metabolism in the dataset to ensure that they were not mistakenly expressed in the diabetic samples. If such gene expressions appeared, it would indicate that something was wrong with the data, necessitating further investigation before proceeding with analysis.

Additionally, sanity checks also involve the inclusion of negative controls, where researchers deliberately include genes or proteins that are not expected to be relevant to the condition being studied. This acts as a safeguard against false positives and misinterpretation of the data. By comparing the dataset to known biological patterns, researchers can detect potential anomalies and address them before proceeding to the next step.

After the sanity checks are completed, the data moves into the data cleaning phase. In this stage, missing data points, outliers, and other anomalies are handled. In vivo datasets, which involve experiments conducted on living organisms, often exhibit natural biological variation. For example, there may be outliers in control groups due to this variation. In such cases, removing outliers might distort the biological reality, so they are often retained in smaller datasets to preserve the integrity of the biological variability.

However, in larger datasets, such as those generated through RNA sequencing or mass spectrometry, true outliers can often be identified and removed without negatively impacting the analysis. It is crucial during this phase to strike a balance between removing erroneous data and preserving meaningful biological variability. In cases where control or treatment groups show large variability, this could reflect true biological responses to a treatment, and removing such data could reduce the accuracy of the analysis.

Once the data has been cleaned, it is then subjected to descriptive analytics. This stage involves generating statistical summaries that reveal patterns, trends, or anomalies in the data. Through methods such as statistical tests, plots, and data visualisation, researchers can begin to interpret the data and form insights. This phase often helps confirm whether the data aligns with the original hypothesis or if new patterns have emerged that warrant further investigation.

The final step is the generation of a report. This document summarises the data analysis process, including the key findings, statistical analyses, and interpretations. Reports are often shared with collaborators or stakeholders and serve as a record of the investigation. Importantly, the report can also raise new scientific questions, leading to further iterations of the workflow. This iterative nature of scientific investigation often leads to new discoveries as researchers refine their understanding of the data.