# Input peak counts vs binarized

**URL:** <https://discourse.scverse.org/t/input-peak-counts-vs-binarized/239>\
**Category:** scvi-tools\
**Tags:** peakvi\
**Created:** [December 22, 2021, 4:05pm UTC](https://discourse.scverse.org/t/input-peak-counts-vs-binarized/239 "2021-12-22T16:05:40Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![bjstewart1](https://yyz1.discourse-cdn.com/flex035/user_avatar/discourse.scverse.org/bjstewart1/32/30_2.png) [@bjstewart1](https://discourse.scverse.org/u/bjstewart1)\
**Post date:** [December 22, 2021, 4:05pm UTC](https://discourse.scverse.org/t/input-peak-counts-vs-binarized/239/1 "2021-12-22T16:05:40Z")

</div>

Do you recommend using counts per peak/region or a binarized matrix as input for peakVI. Apologies in advance if I have missed this somewhere in the documentation.

---

<div class="post-metadata">

**Author:** ![Tal\_Ashuach](https://yyz1.discourse-cdn.com/flex035/user_avatar/discourse.scverse.org/tal_ashuach/32/48_2.png) [@Tal\_Ashuach](https://discourse.scverse.org/u/Tal_Ashuach)\
**Post date:** [December 22, 2021, 11:46pm UTC](https://discourse.scverse.org/t/input-peak-counts-vs-binarized/239/2 "2021-12-22T23:46:02Z")

</div>

You can use either as input, but generally, the counts allow the model to better estimate the library size, so that’s what we recommend.