Model fitting issue with HiFi data

[hae3]

Thank you so much for providing such a great tool!
I am attempting de novo genome assembly using PacBio HiFi data for a diploid plant species, and I used GenomeScope 2.0 for genome size estimation.
From what I can see, the first peak appears to be between 70-80x coverage, and the second peak (I'm not sure if we can call it a peak as it looks quite dispersed) seems to be around 140x. I tried setting the average k-mer coverage for polyploid genome to 70, but didn't see significant differences in the results.
I feel like the model seems to be struggling to fit the observed distribution. Is this affecting the reliability of genome size and heterozygosity estimates? Could high coverage be causing these fitting issues?
Do you have any suggestions or advice I could follow?
Thank you for any guidance!

http://genomescope.org/genomescope2.0/analysis.php?code=Ock3bhuj3IBGvHk0mDc6

[mschatz]

Thanks for your interest. I agree, I dont love the fit here, which is typically due to a combination of error rates and the underlying genome structure. If I adjust some of the parameters I can shift the fit some, but I also dont love it as it assumes a very high rate of heterozygosity (>10%), which is very uncommon: http://genomescope.org/genomescope2/analysis.php?code=idYHep27vOy5nFyGKDVH But you have really nice coverage so I would expect a good assembly. Have you tried hifiasm? This would give some clues on the true genome size and level of heterozygosity (especially if you see many or few duplicates in the BUSCO analysis). And you may want to consider downsampling the data a bit for both GenomeScope and hifiasm - these both work best when you have about ~50x coverage Good luck! Mike

…

On Sun, Feb 23, 2025 at 4:33 AM hae3 ***@***.***> wrote: Thank you so much for providing such a great tool! I am attempting de novo genome assembly using PacBio HiFi data for a diploid plant species, and I used GenomeScope 2.0 for genome size estimation. From what I can see, the first peak appears to be between 70-80x coverage, and the second peak (I'm not sure if we can call it a peak as it looks quite dispersed) seems to be around 140x. I tried setting the average k-mer coverage for polyploid genome to 70, but didn't see significant differences in the results. I feel like the model seems to be struggling to fit the observed distribution. Is this affecting the reliability of genome size and heterozygosity estimates? Could high coverage be causing these fitting issues? Do you have any suggestions or advice I could follow? Thank you for any guidance! transformed_linear_plot.png (view on web) <https://github.com/user-attachments/assets/e641fe18-5eca-4f57-9f08-8f153001e311> transformed_log_plot.png (view on web) <https://github.com/user-attachments/assets/5b595539-b874-4bf3-aea8-ed6b39f8a687> http://genomescope.org/genomescope2.0/analysis.php?code=Ock3bhuj3IBGvHk0mDc6 — Reply to this email directly, view it on GitHub <#154>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AABP34YABMROB3TIVC4J43L2RGIYPAVCNFSM6AAAAABXWDF24CVHI2DSMVQWIX3LMV43ASLTON2WKOZSHA3TGMBVGQ4DSMA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***> [image: hae3]*hae3* created an issue (schatzlab/genomescope#154) <#154> Thank you so much for providing such a great tool! I am attempting de novo genome assembly using PacBio HiFi data for a diploid plant species, and I used GenomeScope 2.0 for genome size estimation. From what I can see, the first peak appears to be between 70-80x coverage, and the second peak (I'm not sure if we can call it a peak as it looks quite dispersed) seems to be around 140x. I tried setting the average k-mer coverage for polyploid genome to 70, but didn't see significant differences in the results. I feel like the model seems to be struggling to fit the observed distribution. Is this affecting the reliability of genome size and heterozygosity estimates? Could high coverage be causing these fitting issues? Do you have any suggestions or advice I could follow? Thank you for any guidance! transformed_linear_plot.png (view on web) <https://github.com/user-attachments/assets/e641fe18-5eca-4f57-9f08-8f153001e311> transformed_log_plot.png (view on web) <https://github.com/user-attachments/assets/5b595539-b874-4bf3-aea8-ed6b39f8a687> http://genomescope.org/genomescope2.0/analysis.php?code=Ock3bhuj3IBGvHk0mDc6 — Reply to this email directly, view it on GitHub <#154>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AABP34YABMROB3TIVC4J43L2RGIYPAVCNFSM6AAAAABXWDF24CVHI2DSMVQWIX3LMV43ASLTON2WKOZSHA3TGMBVGQ4DSMA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[hae3]

Hello Mike,

Thanks so much for your thoughtful feedback. It’s helpful to hear your perspective.

I’ve run an assembly with hifiasm and got around 1 Gb in total assembly size. However, a related species in the same genus has an estimated genome size of around 2.4 Gb (not from my own measurements), so I’m a bit unsure whether the current assembly fully captures the genome. The BUSCO score is C:95.5% [S:89.4%, D:6.1%], based on the eudicots_odb10 dataset.

I'll try downsampling to ~50x, as you suggested, and rerun both GenomeScope and hifiasm to see if the fit improves or if I can get a better size estimate.

Really appreciate your advice — thanks again!

[mschatz]

Those are good busco scores so would expect it to be a relatively complete assembly. Do you have access to the data from the related species? Maybe there is an issue there Mike

…

On Thu, Apr 17, 2025 at 6:27 PM hae3 ***@***.***> wrote: Hello Mike, Thanks so much for your thoughtful feedback. It’s helpful to hear your perspective. I’ve run an assembly with hifiasm and got around 1 Gb in total assembly size. However, a related species in the same genus has an estimated genome size of around 2.4 Gb (not from my own measurements), so I’m a bit unsure whether the current assembly fully captures the genome. The BUSCO score is C:95.5% [S:89.4%, D:6.1%], based on the eudicots_odb10 dataset. I'll try downsampling to ~50x, as you suggested, and rerun both GenomeScope and hifiasm to see if the fit improves or if I can get a better size estimate. Really appreciate your advice — thanks again! — Reply to this email directly, view it on GitHub <#154 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AABP34ZQEDJL3CSLKMIZJH322ATGHAVCNFSM6AAAAABXWDF24CVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDQMJUGEZTEOJRHE> . You are receiving this because you commented.Message ID: ***@***.***> *hae3* left a comment (schatzlab/genomescope#154) <#154 (comment)> Hello Mike, Thanks so much for your thoughtful feedback. It’s helpful to hear your perspective. I’ve run an assembly with hifiasm and got around 1 Gb in total assembly size. However, a related species in the same genus has an estimated genome size of around 2.4 Gb (not from my own measurements), so I’m a bit unsure whether the current assembly fully captures the genome. The BUSCO score is C:95.5% [S:89.4%, D:6.1%], based on the eudicots_odb10 dataset. I'll try downsampling to ~50x, as you suggested, and rerun both GenomeScope and hifiasm to see if the fit improves or if I can get a better size estimate. Really appreciate your advice — thanks again! — Reply to this email directly, view it on GitHub <#154 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AABP34ZQEDJL3CSLKMIZJH322ATGHAVCNFSM6AAAAABXWDF24CVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDQMJUGEZTEOJRHE> . You are receiving this because you commented.Message ID: ***@***.***>