Irina Pochinok
@irinapochinok.bsky.social
doing stuff, HanganuOpatzLab
7/7
We implemented iSTTC to tackle challenges in our own sparse, developmental datasets, and it’s working beautifully. We hope it helps others working with tricky spiking data. Try it yourself github.com/iinnpp/isttc!
We implemented iSTTC to tackle challenges in our own sparse, developmental datasets, and it’s working beautifully. We hope it helps others working with tricky spiking data. Try it yourself github.com/iinnpp/isttc!
GitHub - iinnpp/isttc: iSTTC: intrinsic neural timescales estimation
iSTTC: intrinsic neural timescales estimation. Contribute to iinnpp/isttc development by creating an account on GitHub.
github.com
August 5, 2025 at 8:33 AM
7/7
We implemented iSTTC to tackle challenges in our own sparse, developmental datasets, and it’s working beautifully. We hope it helps others working with tricky spiking data. Try it yourself github.com/iinnpp/isttc!
We implemented iSTTC to tackle challenges in our own sparse, developmental datasets, and it’s working beautifully. We hope it helps others working with tricky spiking data. Try it yourself github.com/iinnpp/isttc!
6/7
iSTTC doesn’t just perform well in simulations; it works in the mess of real neural data, too. Using 30 min of Neuropixels recordings from the Visual Coding @alleninstitute.org dataset, iSTTC gave more stable, more accurate, and more inclusive IT estimates than other methods.
iSTTC doesn’t just perform well in simulations; it works in the mess of real neural data, too. Using 30 min of Neuropixels recordings from the Visual Coding @alleninstitute.org dataset, iSTTC gave more stable, more accurate, and more inclusive IT estimates than other methods.
August 5, 2025 at 8:32 AM
6/7
iSTTC doesn’t just perform well in simulations; it works in the mess of real neural data, too. Using 30 min of Neuropixels recordings from the Visual Coding @alleninstitute.org dataset, iSTTC gave more stable, more accurate, and more inclusive IT estimates than other methods.
iSTTC doesn’t just perform well in simulations; it works in the mess of real neural data, too. Using 30 min of Neuropixels recordings from the Visual Coding @alleninstitute.org dataset, iSTTC gave more stable, more accurate, and more inclusive IT estimates than other methods.
5/7
We didn’t set out to show this, but... 🫢ITs estimated from epoched spike data are dramatically less reliable, with up to 10x more estimation error than continuous data. Regardless of the method, this instability is real. iSTTC helps, but long and uninterrupted recordings still matter a lot.
We didn’t set out to show this, but... 🫢ITs estimated from epoched spike data are dramatically less reliable, with up to 10x more estimation error than continuous data. Regardless of the method, this instability is real. iSTTC helps, but long and uninterrupted recordings still matter a lot.
August 5, 2025 at 8:32 AM
5/7
We didn’t set out to show this, but... 🫢ITs estimated from epoched spike data are dramatically less reliable, with up to 10x more estimation error than continuous data. Regardless of the method, this instability is real. iSTTC helps, but long and uninterrupted recordings still matter a lot.
We didn’t set out to show this, but... 🫢ITs estimated from epoched spike data are dramatically less reliable, with up to 10x more estimation error than continuous data. Regardless of the method, this instability is real. iSTTC helps, but long and uninterrupted recordings still matter a lot.
4/7
It also beats PearsonR on epoched data by a wide margin. iSTTC yields ~17% lower estimation error and ~10% fewer failed fits: more accurate and representative IT estimates!
It also beats PearsonR on epoched data by a wide margin. iSTTC yields ~17% lower estimation error and ~10% fewer failed fits: more accurate and representative IT estimates!
August 5, 2025 at 8:31 AM
4/7
It also beats PearsonR on epoched data by a wide margin. iSTTC yields ~17% lower estimation error and ~10% fewer failed fits: more accurate and representative IT estimates!
It also beats PearsonR on epoched data by a wide margin. iSTTC yields ~17% lower estimation error and ~10% fewer failed fits: more accurate and representative IT estimates!
3/7
iSTTC outperforms classic autocorrelation (ACF) on synthetic continuous data, especially under low firing rates and high burstiness.
iSTTC outperforms classic autocorrelation (ACF) on synthetic continuous data, especially under low firing rates and high burstiness.
August 5, 2025 at 8:29 AM
3/7
iSTTC outperforms classic autocorrelation (ACF) on synthetic continuous data, especially under low firing rates and high burstiness.
iSTTC outperforms classic autocorrelation (ACF) on synthetic continuous data, especially under low firing rates and high burstiness.
2/7
ITs are estimated both on continuous and epoched data, but with inconsistent methods (ACF vs Pearson’s R). iSTTC fixes this: the same algorithm works on both data types!
ITs are estimated both on continuous and epoched data, but with inconsistent methods (ACF vs Pearson’s R). iSTTC fixes this: the same algorithm works on both data types!
August 5, 2025 at 8:28 AM
2/7
ITs are estimated both on continuous and epoched data, but with inconsistent methods (ACF vs Pearson’s R). iSTTC fixes this: the same algorithm works on both data types!
ITs are estimated both on continuous and epoched data, but with inconsistent methods (ACF vs Pearson’s R). iSTTC fixes this: the same algorithm works on both data types!
1/7
Why does this matter? ITs tell us how neurons integrate information over time, a critical link between neural dynamics and cognition. But current methods suffer from bias and limited applicability, especially in biologically realistic conditions (low firing rates and high burstiness).
Why does this matter? ITs tell us how neurons integrate information over time, a critical link between neural dynamics and cognition. But current methods suffer from bias and limited applicability, especially in biologically realistic conditions (low firing rates and high burstiness).
August 5, 2025 at 8:28 AM
1/7
Why does this matter? ITs tell us how neurons integrate information over time, a critical link between neural dynamics and cognition. But current methods suffer from bias and limited applicability, especially in biologically realistic conditions (low firing rates and high burstiness).
Why does this matter? ITs tell us how neurons integrate information over time, a critical link between neural dynamics and cognition. But current methods suffer from bias and limited applicability, especially in biologically realistic conditions (low firing rates and high burstiness).