REF and ALT tracks look identical even with high raw_score and quantile_score

Help & Support
1

Hello,

I am analyzing variant effects using AlphaGenome and I have a question about how to interpret the visualization results.

When I plot the predictions for REF and ALT sequences, the curves often appear visually overlapping in the figure. However, when I compute the numerical difference (ALT − REF), there is a non-zero effect.

For example:
max_delta = 8

In addition, the results from the variant scoring step show both a high raw_score and a high quantile_score.

In my analysis I also filtered variants using a quantile_score threshold above 0.9 (a threshold defined for my study). However, even among these variants with high quantile_score and high raw_score, the RNA-seq plots sometimes still look almost identical.

This raises a few questions:

  1. Is it expected that REF and ALT tracks may appear visually identical even for variants with high raw_score and quantile_score (>0.9)?

  2. In other words, can a variant have a high predicted impact according to the scoring metrics, but still produce changes that are difficult to visually detect in the RNA-seq plots?

  3. Should interpretation of the variant effect rely mainly on the raw_score (or aggregated score) rather than the visual difference in the plotted tracks?

My understanding is that the quantile_score represents the rank of the raw_score relative to a background distribution of common variants, indicating how extreme the predicted effect is, rather than how large the visual difference in the track should appear.

I would appreciate any clarification on how to interpret this situation.

Thank you!

HLA-C
HLA-C1603×597 88.9 KB

2

Hi there!

Thanks for reaching out.

If your REF and ALT tracks appear visually similar despite high differences inraw_score it is likely that plot scaling is masking the difference. Try reducing the interval within plot_components to zoom in along the X axis for better visibility of differences.

It is also possible to have small change in raw_score between REF and ALT despite a high quantile_score. The quantile_score represents the percentile rank of the raw_score compared to a background distribution of common variants, standardizing the predicted impact across different scales rather than reflecting visual magnitude. So even if the change in raw_score is small in absolute numbers, it may be extremely unusual and significant compared to typical background noise and therefore yield a high absolute quantile_score.

Interpretation should primarily rely on the calculated variant scores rather than visual differences.

Kind regards,

Nicolene