Five morphology measures increased threshold-contrast prediction error in 111 Allen human spiny cells at ridge penalty 10

We tested whether five released morphology measures improve prediction of ramp minus long-square spike threshold voltage after electrical properties and cell labels are known. We joined Allen Cell Types tables by cell ID and kept donors separate in five-fold tests with ten partitions. Human: 111 cells from 31 donors. Root mean squared error changed from 2.203 to 2.274 mV. The reduction in squared error was -6.54%, with a conditional 95% interval from -10.89% to -1.77%. Mouse: 233 cells from 172 donors. Root mean squared error changed from 2.219 to 2.255 mV. The reduction in squared error was -3.27%, with a conditional 95% interval from -9.30% to 2.37%. These are selected slice cells and stored protocol summaries. The intervals keep fitted errors fixed. The test does not establish a causal mechanism or transfer across species. Code, input hashes, derived errors, and checks are supplied.

Allen Institute · morphology · spike threshold · donor validation · structure and function