Start with a question you can answer
Research preparation means learning to turn curiosity into a clear, manageable question. The habits below are useful for MPhil and MRes students working with experiments, computational analyses or literature. This optional activity takes about 35–50 minutes; use it to practise thinking, rather than to prepare a finished project proposal.
Begin with an interest, then specify the system, comparison and outcome. “I am interested in gene regulation” is a topic. “In this defined model, does a short stress exposure change the measured expression of gene X compared with an untreated control?” is a question that can guide a design.
Check feasibility before adding complexity
Ask what evidence would answer the question, whether suitable data or materials are available, which skills you would need, and what can be completed within the available time. Include time for checking results and writing. A smaller question with a clear comparison is usually easier to interpret than a broad question with many uncontrolled differences.
| Approach | Possible focus | Early feasibility check |
|---|---|---|
| Experimental study | Measure one defined response to a controlled intervention. | Can the comparison and independent replication be achieved? |
| Computational study | Evaluate a method using a clearly defined benchmark. | Are appropriate input data, expected outputs and computing resources available? |
| Literature investigation | Compare evidence for a focused biological question. | Can studies be identified and compared using explicit inclusion criteria? |
These are general approaches, not a list of course project options. Actual project scope and arrangements are developed through the course and supervision.
Know what your design can establish
In an observational study, researchers measure existing differences without assigning the exposure of interest. In an experiment, they deliberately apply an intervention. An association may reflect other differences between groups: a confounder can make an apparent relationship misleading. Randomisation helps distribute background differences in an experiment; it does not guarantee perfect balance in a small sample.
A batch effect is systematic variation associated with how or when samples are processed. If every control is processed on Monday and every treated sample on Tuesday, treatment and day are confounded. Statistical adjustment cannot reliably separate two effects that never vary independently in the data.
Where feasible, spread comparison groups across processing batches and record the batch information. The NC3Rs guide explains nuisance variables, blocking and confounding; although its examples concern animal experiments, the design questions are useful more broadly.
A fictional design to discuss
Imagine twelve independently grown yeast cultures: six are randomly assigned to a stress exposure and six to a control condition. Each culture produces one biological sample. The numbers below illustrate the design; they are not a sample-size recommendation.
| Processing batch | Control cultures | Stress-exposed cultures |
|---|---|---|
| A | 3 | 3 |
| B | 3 | 3 |
The measured response is expression of a selected gene. Both groups appear in both batches, and processing order can be randomised within each batch. Before seeing the results, define the outcome, relevant quality checks and any justified exclusions. Record deviations when they occur.
Biological and technical replication
Independent cultures provide biological replication for this question. Repeating a measurement twice from the same sample provides technical replication: it helps assess measurement consistency but creates no extra independent cultures. Here there are twelve biological samples even if there are twenty-four assay readings.
Always identify the unit that can receive an intervention independently and the population to which you want to generalise. Independence depends on the design; counting tubes, reads or cells is insufficient. See the NC3Rs explanation of the experimental unit.
Would ten extra measurements of each existing sample solve a shortage of independent cultures?
No. More technical repeats may improve measurement precision, but they do not reveal the biological variation that additional independent cultures would capture.
Leave a record that explains your result
Keep original inputs separate from derived results. Use meaningful filenames and retain earlier versions when an analysis changes. Cambridge provides practical guidance on file organisation and version control and documenting research data.
A useful notebook entry records:
- The date, question and purpose of the work.
- Input filenames, source, version, and permission or licence for use.
- Sample identifiers, variable definitions, units and relevant batches.
- The method or code version, software, parameters and reference assembly where relevant.
- Quality checks, exclusions, changes of plan and their reasons.
- Output filenames, observations, interpretation and the next question.
For literature work, record databases, search dates and terms, inclusion decisions, and where each extracted result came from. Someone should be able to trace your conclusion back to its evidence.
Your first practice plan
Choose a fictional question and write five lines: the question, the comparison, the outcome, the main alternative explanation, and the record you would need to keep. Then identify one obstacle that could make the question impractical.
What should you do if the available evidence cannot answer your original question?
Revise the question or design explicitly. Describe what the evidence can support, preserve the record of the change, and avoid presenting an exploratory finding as a prediction made before the analysis.
Updated September 2026.