infer_from_cds in UTRFetcher does not work correctly

[Florian R. Hölzlwimmer]

Currently, UTR region inference works only for non-spliced UTR regions:

kipoiseq/kipoiseq/extractors/gtf.py

Lines 348 to 368 in e67fab6

 # get start and end of cds for each transcript 
 cds = CDSFetcher.get_cds_from_gtf(df=df, on_error_warn=on_error_warn) \ 
     .groupby('transcript_id') \ 
     .agg({'Start': min, 'End': max}) 
  
 # join cds start and end to utr df 
 utr_df = df.query("Feature == 'transcript'") \ 
     .set_index('transcript_id') \ 
     .join(cds, rsuffix="_cds") \ 
     .dropna(subset=['Start_cds', 'End_cds'], axis=0) 
  
 if feature_type.upper() == "5UTR": 
     utr_df['Start'] = np.where(utr_df['Strand'] == '+', int(utr_df['Start']), int(utr_df['End_cds'])) 
     utr_df['End'] = np.where(utr_df['Strand'] == '+', int(utr_df['Start_cds']), int(utr_df['End'])) 
     utr_df['Feature'] = pd.Categorical("5UTR", categories = utr_df['Feature']) 
 if feature_type.upper() == "3UTR": 
     utr_df['Start'] = np.where(utr_df['Strand'] == '+', int(utr_df['End_cds']), int(utr_df['Start'])) 
     utr_df['End'] = np.where(utr_df['Strand'] == '+', int(utr_df['End']), int(utr_df['Start_cds'])) 
     utr_df['Feature'] = pd.Categorical("3UTR", categories = utr_df['Feature']) 
  
 utr_df.drop(['Start_cds', 'End_cds'], axis=1, inplace=True) 

TODO:

1. Generate test data with transcript that contains a spliced UTR, e.g.:
   tabix /s/genomes/GenBank/hg38/annotation/hg38.ensGene.gtf.gz chr22 | grep -i ENST00000263207 > kipoiseq/tests/data/chr22_ENST00000263207.gtf
2. Simulate some variants:

• chr22_ENST00000263207_3UTR.vcf.gz
• chr22_ENST00000263207_5UTR.vcf.gz

3. Generate with infer_from_cds=False:

• chr22_ENST00000263207_3UTR.alt_seqs.txt
• chr22_ENST00000263207_3UTR.ref_seq.txt
• chr22_ENST00000263207_5UTR.alt_seqs.txt
• chr22_ENST00000263207_5UTR.ref_seq.txt

4. Update tests:

   kipoiseq/tests/extractors/test_protein.py

   Lines 323 to 358 in 1d72daf

    # chr22_fasta_file = 'tests/data/chr22.fa.gz' 
    chr22_gtf_file = 'tests/data/chr22_ENST00000319363.gtf' 
     
     
    # chr22_5UTR_vcf_file = 'tests/data/chr22_ENST00000319363_5UTR.vcf.gz' 
     
     
    def test_5UTRFetcher__read_utr(): 
        utr5 = UTRFetcher._read_utr(chr22_gtf_file, feature_type="5UTR") 
     
        assert utr5.shape == (1, 12) 
     
        assert utr5.iloc[0].Chromosome == 'chr22' 
        assert utr5.iloc[0].Start == 17565848 
        assert utr5.iloc[0].End == 17565981 
        assert utr5.iloc[0].Strand == "+" 
     
        utr5_from_cds = UTRFetcher._read_utr(chr22_gtf_file, feature_type="5UTR", infer_from_cds=True) 
     
        pd.testing.assert_frame_equal(left = utr5.drop(['exon_number', 'exon_id'], axis=1), right = utr5_from_cds.drop(['exon_number', 'exon_id'], axis=1), check_dtype=False) 
     
     
    def test_3UTRFetcher__read_utr(): 
        utr3 = UTRFetcher._read_utr(chr22_gtf_file, feature_type="3UTR") 
     
        assert utr3.shape == (1, 12) 
     
        assert utr3.iloc[0].Chromosome == 'chr22' 
        assert utr3.iloc[0].Start == 17590710 
        assert utr3.iloc[0].End == 17596583 
        assert utr3.iloc[0].Strand == "+" 
     
        utr3_from_cds = UTRFetcher._read_utr(chr22_gtf_file, feature_type="3UTR", infer_from_cds=True) 
     
        pd.testing.assert_frame_equal(left=utr3.drop(['exon_number', 'exon_id'], axis=1), 
                                      right=utr3_from_cds.drop(['exon_number', 'exon_id'], axis=1), check_dtype=False) 

[Florian R. Hölzlwimmer]

The correct solution would be to pyranges.intersect() the exons with transcript_start - CDS_start for 5'UTR and CDS_end - transcript_end for 3'UTR.
Also, make sure that the last / first exon of the UTRs is shrinked to CDS_start / CDS_end.