AI Is Transforming Spike Sorting — But It Won't Replace the Electrophysiologist

A neuron fires an electrical signal in 1 millisecond.

Now imagine sorting thousands of them. Across hundreds of electrodes. Simultaneously. And every signal looks almost like the next one.

That's been my job for 12 years.

The problem: untangling the chaos

In electrophysiology, we record the electrical activity of neurons to test drug candidates for the pharmaceutical industry. The raw data is a chaos of overlapping signals. You have to untangle them, assign each one to the right neuron, and extract usable pharmacological parameters.

This is called spike sorting, and it's one of the hardest problems in biological signal processing.

Why I'm bringing this up now

Because AI is transforming the field. In just a few months:

• Kilosort4 (Nature Methods) introduced graph-based clustering to automatically sort spikes across thousands of channels
• SimSort (NeurIPS 2025) showed that a model trained only on simulated data can generalize zero-shot to real recordings
• SpikeInterface (GitHub) offers a modular open-source framework that lets you compare and combine the best algorithms in a few lines of Python

What used to take days of manual sorting now takes minutes.

The reality check no one talks about

AI doesn't replace the electrophysiologist. It augments them.

An algorithm can sort 10,000 spikes. It can't tell you whether the pharmacological profile makes sense, whether the slice is viable, or whether series resistance drifted during the recording.

That's exactly where dual expertise makes the difference: understanding neuroscience AND being able to code the analysis pipelines.

What this means for biotech and CROs

12 years of electrophysiology + data science. PhD in Neuroscience. PharmD. Today, I design AI pipelines that automate this kind of analysis for the pharmaceutical industry.

If you work in biotech or a CRO and your electrophysiology data is still being sorted by hand, there are other options — faster, more reproducible, and without sacrificing scientific rigor.

If you'd like to discuss it, book a 30-minute discovery call. We'll look together at what can be automated in your current pipeline.

---