A Multispectral Sensor Hacked Into A qPCR Machine...

UPDATE: There’s a follow up post where I acquired useful data from this system.

A while back I wrote about CoDx. One of the things I found interesting about their approach was the suggestion that a multispectral sensor could be used in a qPCR machine. qPCR machines already routinely use very cheap ~$1 light-to-voltage sensors. But a multi-spectral sensor potentially means replacing a somewhat complex optical system with a single part.

Here’s the optical detection from a cepheid smart cycler 2:

The multi-spectral sensor provides light sensing across 9 different wavelengths in a cheap (~$10) package, potentially replacing all of that optics above:

So I figured why not try and throw out all that optical “stuff” and see what happens if we stick this cheap sensor in there instead:

And then wire the whole thing up to a Raspberry Pi and stick the module back in the qPCR machine:

This means we can still run the standard protocols and use the qPCR machine excitation system, but we can see how the detection system performs. Comparing the excitation wavelengths of the SmartCycler and FAM,Cy3,Cy5, Texas Red against the detection wavelengths of this sensor we get the following:

I’ve shown the point at which the relative sensitivity of each sensor drops below 10% and picked the best sensors for each dye where they don’t overlap with the SmartCycler excitation wave lengths. Emissions are solid, dotted lines are excitation wavelengths for the named dye.

The above suggests I want to be looking at the 555nm,630nm and 660nm sensors. If do a run on the SmartCycler and record the detection events I get the following plot:

caption...

This appears to me to be showing somewhat reasonable rejection for the excitation (the 555nm/FAM detection wavelength stays reasonably low when the excitation for channel 1 is on etc.). There’s actually just an old PCR tube sitting in the cycler, so it’s possible that we might be seeing some real fluorescence.

We’ve also seeing bleed through where we would expect to (for example between C3 565nm to 590nm and the 555nm detection channel). Which suggests things are working more or less as expected.

Next of course I should probably try a real sample, consider subscribing if you’d like to support that…