Training Topics on iBOTS Group - Research Software Engineering Training

Developer Environment

Mon, 01 Jan 0001 00:00:00 +0000

Computational research projects bring together code, data, documentation, and software dependencies. Keeping these components organized makes projects easier to understand, share with collaborators, and revisit later.

In this topic, you will learn how several essential tools support different parts of that work: Visual Studio Code for writing and organizing code; Git and GitHub for version control and project management; Sciebo for sharing data and other project materials; and the terminal and package managers for creating reproducible computational environments.

Programming and Data Acquisition

Mon, 01 Jan 0001 00:00:00 +0000

Experimental data acquisition connects software to physical measurements. Arduino microcontrollers can read sensors and input devices, record task events, and provide useful timestamps for later analysis. The structure and reliability of the code therefore directly affect the quality and interpretability of the collected data.

Why This Matters

Experimental code sits close to the data itself. A small timing bug, missing metadata field, or unstable signal can make downstream analysis harder to trust. Arduino provides an accessible entry point for learning how software, sensors, timestamps, and physical measurements fit together. This topic presents Arduino as both a maker skill and a way to reason about reliable data acquisition.

Local Field Potential (LFP) Analysis

Mon, 01 Jan 0001 00:00:00 +0000

Local field potentials (LFPs) are extracellular voltage signals generated by the combined transmembrane currents of neuronal populations. Their interpretation depends on understanding how recording conditions, preprocessing choices, and nearby neural processes shape the signal. This topic connects LFP analysis with spike and behavioral data.

Why This Matters

Reliable LFP analysis requires attention to signal quality, neurophysiological interpretation, and reproducible computation. Filtering choices, line noise, synchronization, reference schemes, and behavioral alignment all influence power spectra and coherence estimates. This topic connects analytical methods with the practical checks needed to interpret results with confidence.

Calcium Imaging Analysis

Mon, 01 Jan 0001 00:00:00 +0000

Calcium imaging produces movies in which changes in fluorescence provide an indirect measure of neural activity. Transforming raw movies into interpretable signals involves motion correction, region selection, trace extraction, neuropil correction, and careful interpretation of the relationship between fluorescence and neural spiking.

Why This Matters:

Calcium imaging analysis involves a chain of interdependent decisions, including file conversion, downsampling, motion correction, segmentation, neuropil correction, trace extraction, and quality control. Each decision affects subsequent results and what can be interpreted from them. This topic helps you understand the full workflow, compare open-source tools, and document analysis choices so that results remain traceable from raw movies to final figures.

Neural Spike Analysis

Mon, 01 Jan 0001 00:00:00 +0000

How do we extract, measure, and model spike trains from neuronal recordings? Using real and simulated electrophysiological data, you will investigate the firing patterns of individual neurons and neural populations. You will explore descriptive techniques such as raster plots, peristimulus time histograms (PSTHs), and cross-correlograms, alongside statistical approaches including Poisson-based generative models and coherence-based analyses.

Why This Matters

Spike analysis turns raw voltage traces into interpretable measurements of neural activity, so every step matters: filtering, detection, sorting, quality control, alignment, summary statistics, and modeling. Early processing decisions affect raster plots, PSTHs, cross-correlograms, and population-level interpretations. This topic connects an intuitive understanding of the methods with reproducible Python workflows.

(M)EEG Analysis

Mon, 01 Jan 0001 00:00:00 +0000

EEG and MEG provide non-invasive views of neural activity with millisecond-scale temporal resolution. EEG measures electrical potential differences at the scalp, whereas MEG measures magnetic fields associated with neural currents. In both modalities, each sensor captures a mixture of activity from multiple neural sources, making the relationship between recorded signals and underlying processes an important part of the analysis.

Why This Matters

Analyzing EEG and MEG data requires careful alignment of experimental events, sensor recordings, preprocessing decisions, and statistical interpretations. MNE-Python provides a powerful open-source toolkit, but scientifically sound conclusions depend on understanding how choices about artifact handling, epoching, time-frequency analysis, source separation, connectivity, and decoding affect the results. This topic connects these concepts and theoretical foundations to transparent, reproducible computational workflows.

Website Development

Mon, 01 Jan 0001 00:00:00 +0000

A personal research website can bring together projects, publications, a CV, blog posts, figures, and other scientific outputs. Separating this content from the site layout makes the website easier to update and maintain as the work develops.

Why This Matters

Research websites can do more than list publications. They can make methods easier to find, document software for collaborators, share teaching materials, and connect notebooks or figures to the story behind a project. This topic emphasizes websites that can be maintained with ordinary research tools: Markdown, Git, reusable templates, and static hosting.

Machine Learning

Mon, 01 Jan 0001 00:00:00 +0000

Machine learning can reveal linear and nonlinear relationships among neural measurements, experimental conditions, and behavior. Drawing reliable conclusions requires more than fitting a model: data representation, normalization, validation design, model comparison, and diagnostic checks all influence whether apparent performance generalizes to new data.

Why This Matters

Machine learning is most useful in research when the path from data preparation to model comparison is clear. Normalization choices, cross-validation design, performance baselines, and diagnostic checks all affect whether a model generalizes beyond its training data. This topic connects these foundations to neural network modeling while emphasizing reproducible code and healthy skepticism about model performance.

AI-Assisted Software Development

Mon, 01 Jan 0001 00:00:00 +0000

Agentic AI tools can help implement, revise, and test research software. Their use also makes it important to define what the software should do, decide which changes are valuable, and establish evidence that the results meet scientific and technical requirements. This topic explores how to organize that work so a project’s development remains understandable and reviewable as the pace of implementation increases.

You will set up coding agents and their project instructions, turn research needs into software specifications, and express expected behavior through behavior-driven development (BDD) scenarios. You will combine unit, integration, acceptance, and regression tests with checks such as Ruff and complexipy, and explore property-based and mutation testing to assess different aspects of correctness and test effectiveness.

Robust and Reusable Software

Mon, 01 Jan 0001 00:00:00 +0000

Scientific code often begins as a script written to answer one question. As it is reused, extended, or shared, its structure and assumptions become increasingly important. Small functions, clear project organization, automated tests, and explicit data requirements help researchers understand what the code does and change it with greater confidence.

Why This Matters

Research software needs to support changing questions, datasets, and collaborators. Clear boundaries and small, understandable components make those changes easier to manage. Tests and validation make expectations visible and help detect problems before they propagate into an analysis.

Automating Research Workflows

Mon, 01 Jan 0001 00:00:00 +0000

Scientific analyses often involve repeating operations across files, choosing steps based on metadata, and running calculations in a particular order. As a project grows, expressing repetition, decisions, and dependencies explicitly helps an analysis remain understandable and repeatable.

Why This Matters

A research workflow contains knowledge that can easily remain implicit: which files belong together, which parameters were used, which steps must run first, and how a result was produced. Making these decisions explicit reduces reliance on memory and manual coordination, helping both you and your collaborators understand and repeat the analysis.

Dataframes and Statistics

Mon, 01 Jan 0001 00:00:00 +0000

Much of scientific analysis begins with tables: trial outcomes, spike times, participant information, and measurements collected across experimental conditions. Dataframes connect these observations with their metadata and support a continuous workflow from data preparation to visualization, statistical modeling, and written interpretation.

Why This Matters

Decisions about rows, columns, data types, grouping, and aggregation determine which observations are compared and what a statistical result represents. Explicit transformations and readable code help collaborators trace analytical decisions, check assumptions, and adapt the workflow when new data arrives.

Arrays and Numerical Computing

Mon, 01 Jan 0001 00:00:00 +0000

Scientific data often takes the form of arrays: recorded signals, images, measurements, and multidimensional datasets. Understanding dimensions, indexing, numerical operations, and visualization helps you turn these measurements into analyses that others can understand and reproduce.

Why This Matters

Array dimensions, indexing choices, and memory behavior affect both the meaning of a result and whether an analysis runs successfully. Clear functions, well-labeled data, version control, and figures that can be regenerated make a numerical workflow easier to maintain and share.