Why pre-screen computational neuroscientists before the technical panel
A model with enough free parameters will fit anything, which is why the useful question is what it predicted that turned out to be wrong. Add the practical problem that much research code cannot be run by anybody except its author, and the screen writes itself. A short screen asks what their model failed to predict and whether somebody else can reproduce the analysis.
What actually matters when screening Computational Neuroscientist candidates
- 01
Theoretical command
Probe command of dynamical systems, Hodgkin-Huxley and integrate-and-fire models, dimensionality reduction (PCA, GPFA), and Bayesian or point-process methods for spike train analysis.
- 02
From theory to hardware or code
Ask what they built: NEURON, Brian2, NEST simulations, PyTorch models fit to Neuropixels or calcium imaging data, and any released code or preprint.
- 03
Research judgement
Test how they chose between competing models, handled overfitting on small trial counts, and decided a hypothesis about circuit function was not worth pursuing further.
- 04
Explaining it to non-specialists
Judge how they explain latent state or attractor findings to wet-lab experimentalists, clinicians, or grant panels who do not read differential equations.
Pre-screening questions to ask Computational Neuroscientist candidates
12 questions grouped by what they test. Ask the same set in every screen and score answers on a consistent scale, or send them as an async video screen and compare answers side by side.
Research they did
3 questions01Tell us about a recent project where you applied computational methods to a neuroscience problem.
Listen forA specific project with the question, method and result explained, and their own contribution clear.
Projects described at abstract level, or contribution to group work left unspecified.
02Can you describe your experience with neural modelling?
Listen forModel classes used with a reason for the choice, and the biological assumptions made explicit.
Models chosen for convenience, or biological plausibility never considered.
03What experience do you have analysing electrophysiological data?
Listen forReal recordings analysed, with spike sorting, artefacts and noise handled from experience.
Only simulated or pre-processed data used, or recording artefacts not recognised.
Models validated
4 questions04How do you approach hypothesis testing in computational experiments?
Listen forPredictions specified before analysis, with statistical approach chosen to suit the data structure.
Hypotheses formed after seeing results, or repeated testing without correction.
05How do you verify the accuracy and reliability of your models?
Listen forHeld-out data and parameter sensitivity both tested, with a failed prediction described honestly.
Validation on fitted data only, or model quality judged by visual similarity to results.
06Have you worked on large-scale neural simulations?
Listen forSimulations run at scale, with numerical stability and computational cost both understood.
Scale described by neuron count alone, or numerical issues never encountered.
07Which statistical methods do you use most in your work?
Listen forMethods explained with their assumptions, chosen for the data rather than by habit.
Standard tests applied to dependent data, or assumptions never checked.
Code others can run
3 questions08Which programming languages do you use for this research?
Listen forFluent in the relevant scientific stack, able to write code others in the group can maintain.
Coding described as a means to an end, or no experience beyond adapting existing scripts.
09How comfortable are you sharing your code and models?
Listen forCode published with documentation and environment specified, so results can be reproduced.
Code never shared, or sharing resisted because it would need tidying first.
10Can you describe troubleshooting a difficult problem in your research code?
Listen forA specific bug found systematically, including one that had affected published or draft results.
Debugging described as trial and error, or no serious bug ever encountered.
Works with experimentalists
2 questions11How do you integrate experimental data with computational models?
Listen forExperimental constraints and measurement limits understood, with the data quality assessed first.
Data treated as ground truth, or experimental limitations never discussed with the group.
12What interdisciplinary collaborations have you been part of?
Listen forGenuine joint work where experiments changed the model or the model shaped an experiment.
Collaboration described as receiving datasets, or no influence in either direction.
How to score responses
Score every candidate on the same four criteria immediately after the screen. At this stage you are shortlisting for panel interviews, not making the final call.
Theoretical command
35%5Derives model behaviour from first principles, states assumptions behind each method, and names where the theory breaks against real cortical data.
From theory to hardware or code
30%5Points to versioned repositories, simulations that reproduce published results, and pipelines other labs actually ran on their own recordings.
Research judgement
20%5Describes abandoned model families with reasons, uses cross-validation and surrogate data honestly, and separates fit quality from mechanistic claims.
Explaining it to non-specialists
15%5Translates model predictions into testable experiments a slice physiologist could run, without hiding behind notation or overselling correlation.
A model with enough parameters fits anything. A one-way video screen asks what theirs failed to predict.
Try it on HirevireScreening FAQ
Process basics
How long should a pre-screening round for this role take?
Fifteen minutes across eight to ten questions, answered async. Enough to establish research they did, test their modelling rigour, and hear how they work with experimental colleagues.
Should I ask for code as well?
Yes, after the screen. A repository shows documentation, structure and whether the analysis can be rerun, and none of that appears in a publication list.
Evaluating answers
What is the strongest signal when screening this role?
A prediction their model got wrong. Rigorous researchers test against held-out data and report failures. Anyone whose models always fitted has been describing data rather than explaining it.
How do I judge their engineering ability?
Ask whether their code is shared and how a colleague would run it. Real answers describe version control and documented environments. Scripts on a personal machine will not survive their departure.
Go deeper on this role
Sanat has been hiring since 2012 and watching the recruitment industry change up close ever since, and turned that screening process into Hirevire's video screening platform. LinkedIn
