How to visualize predictions made with predict_sequence?

Hi,
I’m a researcher working on DNA regulation, and we generated and inserted some DNA sequences around the AXIN2 gene.
I’d like to run the predictions of RNA-seq signal for our sequences, however I struggle to use the output of predict_sequence.

Here is my code (an extract from the relevant part):

output = dna_model.predict_sequence(
    sequence=seq_string.center(dna_client.SEQUENCE_LENGTH_500KB, 'N'),  # Pad to valid sequence length.
    organism=dna_client.Organism.HOMO_SAPIENS,
    requested_outputs=[dna_client.OutputType.RNA_SEQ],
    ontology_terms=['CL:0000236'],  # B-cell
    #interval=myinterval
)
print(f'RNA-SEQ predictions shape: {output.rna_seq.values.shape}')

seq_string is a DNA sequence of 393216 bp that I pad to the ~500kb length of the model

The prediction works correctly, but I’d like to visualize the output as a track, and I’m not sure how to do that.
Basically the plot_components.plot function fails because it does not know the Interval in the human genome, but:

  • I tried to add a corresponding interval in the predict_sequence() function (as mentionned in your docs, but it does not work: alphagenome.models.dna_client.DnaClient — AlphaGenome
  • I tried to “make a fake interval” but apparently the output.rna_seq.interval field is immutable
  • I saw the predict_variant alternative, and I really like the visualization output, but it does not apply in my case because the sequences are too different (and do not really correspond to different ‘variants’, the way they are usually defined)

How should I proceed?

Thanks in advance

I just figured it out yesterday, you have to install the alphagenome package from the master branch on github - docs must be built from there and then there is an argument interval in predict_sequence

Ho, i see.
And then I will be able to set whether a real genomic coordinate, or to create a fake one only corresponding to my sequence I suppose?
Ok thanks, I’ll try that out!

Yes, you can set anything as the interval as long as the length matches the length of the sequence.

1 Like