Unexpected trend in diversity_fit_mse convergence plot

[zhangming-m]

I am running SINGER on a dataset with ~342 samples. I observed that the diversity_fit_mse (calculated via compute_traces.py) shows an unusual pattern: it initially declines, then rises significantly, and finally stabilizes at a high value 5.4 * 10^-6

this is my SINGER code:
parallel_singer -Ne 2e5 -m 2.69e-7 -vcf /data/zhangmm/linyan/Qv342_chr11.polarized.snps -output Qv342_chr11.parallel_singer -n 800 -polar 0.99 -thin 100 -L 100000 -num_cores 50 -ratio 0.037

Questions:

1. Is this "decline-then-rise" pattern expected for large sample sizes or specific demographic scenarios?
2. The stable MSE value 5.4 * 10^-6 seems much higher than the examples in the paper 10^-8. Does this come from mismatch between my -Ne/-m parameters and the observed diversity pi= 0.005?

[YunDeng98]

Thanks for your question. Maybe the most helpful thing is to plot something like Extended Data Figure 4 for the starting sample and the ending sample, so that I can inspect them?

…

On Jan 4, 2026, at 12:26 AM, zhangming-m ***@***.***> wrote: zhangming-m created an issue (popgenmethods/SINGER#38) <#38> I am running SINGER on a dataset with ~342 samples. I observed that the diversity_fit_mse (calculated via compute_traces.py) shows an unusual pattern: it initially declines, then rises significantly, and finally stabilizes at a high value 5.4 * 10^-6 this is my SINGER code: parallel_singer -Ne 2e5 -m 2.69e-7 -vcf /data/zhangmm/linyan/Qv342_chr11.polarized.snps -output Qv342_chr11.parallel_singer -n 800 -polar 0.99 -thin 100 -L 100000 -num_cores 50 -ratio 0.037 Questions: Is this "decline-then-rise" pattern expected for large sample sizes or specific demographic scenarios? The stable MSE value 5.4 * 10^-6 seems much higher than the examples in the paper 10^-8. Does this come from mismatch between my -Ne/-m parameters and the observed diversity pi= 0.005? Qv342_chr11.parallel_singer.thin100.n800.convergence.png (view on web) <https://github.com/user-attachments/assets/41acc5fb-87d7-44d7-8f3b-41d1529b0ca2> — Reply to this email directly, view it on GitHub <#38>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ANATC4G7KR26WU62ARJFPQ34FDFB5AVCNFSM6AAAAACQUILTNGVHI2DSMVQWIX3LMV43ASLTON2WKOZTG43TQOJZGMYTQOI>. You are receiving this because you are subscribed to this thread.

[zhangming-m]

Thanks for reply. I'll try and reply later~

[zhangming-m]

Hi YunDeng, this my result

this is the first start sample SNP based diversity vs branch length based diversity

this is the last sample SNP based diversity vs branch length based diversity