Hi Juntao, Yes, ChimeraX AlphaFold on Google Colab only handles about 1000 amino acids before running out of memory on the 5 year old GPUs (e.g. P100, K80, T4) with 16 Gbytes of GPU memory that they provide. This is a limitation of AlphaFold. Your 1500 amino acids would run fine on an Nvidia RTX 3090 GPU which has 24 Gbytes, or on higher-end GPUs like an Nvidia A40 with 48 Gbytes. These sequence size limits are a significant limitation of AlphaFold. Here is a web page I made showing AlphaFold runs that failed and succeeded and the largest sequences that worked in these tests on Colab was ~1100 residues, on an RTX 3090 it managed 2100 residues and on an A40 it got up to 3700 residues. https://www.rbvi.ucsf.edu/chimerax/data/alphafold-jan2022/afspeed.html Tom

On Feb 16, 2022, at 6:56 AM, Juntao Yu <juntao_yu@g.harvard.edu> wrote:

Hi Tom,

Thank you for your email. I checked the log and realized the failure was indeed because of out of memory on Google Colab. The protein complex is about 1500 amino acids large so probably that is the reason. It’s good to know that it might take lots of memory for alphaFold to calculate large complexes. I will try to run the prediction on our clusters or lab computers with GPU instead then. Thank you.

Best, Juntao