Data Loaders

​

Overview

​

Classes

​

Fasta

from trifid.data.loaders import Fasta

f = Fasta(path=fasta_path, db=db_name)
df_fasta = f.load

• load: Returns a pandas DataFrame with columns id and sequence

​

GFF

from trifid.data.loaders import GFF

g = GFF(path=gff_path, db="gencode")
df_gff = g.load

• load(complete=False): Returns parsed GFF data as pandas DataFrame

​

GTF

from trifid.data.loaders import GTF

gtf = GTF(path=gtf_path, db="gencode")
df_gtf = gtf.load

• load: Returns parsed GTF data as pandas DataFrame

​

Functions

​

load_annotation

from trifid.data.loaders import load_annotation

df = load_annotation(
    filepath="gencode.v38.annotation.gtf.gz",
    db="g"
)

• transcript_id: Transcript identifier (version stripped)
• CCDS: Boolean for Consensus CDS annotation
• StartEnd_NF: Boolean for unconfirmed start/end regions
• RNA_supported: Boolean for RNA-seq support
• basic: Boolean for basic transcript set membership
• NAGNAG: Boolean for alternative acceptor sites
• readthrough: Boolean for readthrough transcripts
• nonsense_mediated_decay: Boolean for NMD transcripts
• level: Annotation confidence level

• GENCODE GTF Format
• GTF Annotation Format

​

load_appris

from trifid.data.loaders import load_appris

df = load_appris("appris_data.appris.txt")

• gene_id, gene_name: Gene identifiers
• transcript_id, translation_id: Transcript and protein IDs
• ccdsid: CCDS identifier
• tsl: Transcript support level (1-5, or 6 for NA)
• length: Protein length
• firestar: Functional residue score
• matador3d: 3D structure score
• corsair: Cross-species conservation score (normalized)
• spade: Domain prediction score
• thump: Transmembrane/signal peptide score
• crash_p, crash_m: Signal peptide and mitochondrial scores
• appris: APPRIS annotation category

• Filters for transcripts with TRANSLATION flag
• Separates crash scores into peptide (crash_p) and mitochondrial (crash_m)
• Normalizes corsair scores (values ≤1.5 set to 0)
• Strips version numbers from transcript IDs

​

load_corsair_alt

from trifid.data.loaders import load_corsair_alt

df = load_corsair_alt("corsair_alt_scores.tsv.gz")

• transcript_id: Transcript identifier (version stripped)
• corsair_alt: Alternative conservation score (NaN filled with 0)

​

load_corsair_alt_exons

from trifid.data.loaders import load_corsair_alt_exons

df = load_corsair_alt_exons("corsair_alt_exons.tsv.gz")

• transcript_id: Transcript identifier
• minexon_corsair_alt: Minimum exon conservation score per transcript

​

load_phylocsf

from trifid.data.loaders import load_phylocsf

df = load_phylocsf("PhyloCSF_scores.tsv.gz")

• transcript_id: Transcript identifier (version stripped)
• ScorePerCodon: Codon substitution frequency score (minimum per transcript)
• RelBranchLength: Branch length estimation (minimum per transcript)
• PhyloCSF_Psi: Alternative CSF representation (minimum per transcript)

• Filters for exons with RelBranchLength > 0.1 and NumCodons > 3
• Aggregates multi-exon scores by taking minimum values
• NMD variants not included in source data

• PhyloCSF Paper
• PhyloCSF GitHub

​

load_qpfam

from trifid.data.loaders import load_qpfam

df = load_qpfam("qpfam_scores.tsv.gz")

• transcript_id: Transcript identifier
• pfam_score: Overall Pfam domain score
• pfam_domains_impact_score: Domain impact quantification
• perc_Damaged_State: Percentage of damaged domains
• perc_Lost_State: Percentage of lost domains
• Lost_residues_pfam: Number of lost Pfam residues
• Gain_residues_pfam: Number of gained Pfam residues

​

load_qsplice

from trifid.data.loaders import load_qsplice

df = load_qsplice("qsplice_scores.tsv.gz")

• transcript_id: Transcript identifier (version stripped)
• RNA2sj: RNA-seq to splice junction mapping score (capped at 1)
• RNA2sj_cds: CDS-specific splice junction score (capped at 1)

• Replaces ”-” with NaN, fills with 0
• Caps values >1 at 1

​

load_reference

from trifid.data.loaders import load_reference

df = load_reference("reference_annotations.tsv.gz")

• transcript_id: Transcript identifier
• ann_type: Annotation category
• transcript_ref: Reference transcript indicator

• Alternative
• Principal
• Alternative.NMD
• Redundant Principal
• Principal Duplication
• Alternative Duplication
• Redundant Alternative
• Principal.RT (readthrough)
• Principal.NMD
• Alternative.RT

​

load_sequences

from trifid.data.loaders import load_sequences

df = load_sequences("gencode.v38.pc_translations.fa.gz")

• transcript_id: Transcript identifier (version stripped)
• sequence: Protein amino acid sequence

• Uses BioPython’s SeqIO parser
• Extracts transcript ID from FASTA header (pipe-delimited)
• Strips version numbers from IDs

​

load_spade

from trifid.data.loaders import load_spade

df = load_spade("appris_method.spade.gtf")

• seqname, source, feature: GTF standard fields
• start, end: Genomic coordinates
• score: Domain score
• strand, frame: Strand and reading frame
• transcript_id, gene_id: Identifiers
• hmm_name: Hidden Markov Model domain name
• evalue: E-value for domain prediction
• pep_start, pep_end: Peptide coordinates

​

Complete Example

from trifid.data.loaders import (
    load_annotation,
    load_appris,
    load_sequences,
    load_qpfam,
    load_phylocsf
)

# Load genome annotations
df_annotation = load_annotation(
    "gencode.v38.annotation.gtf.gz",
    db="g"
)

# Load APPRIS functional scores
df_appris = load_appris("appris_data.appris.txt")

# Load protein sequences
df_sequences = load_sequences("gencode.v38.pc_translations.fa.gz")

# Load domain scores
df_qpfam = load_qpfam("qpfam_scores.tsv.gz")

# Load evolutionary scores
df_phylocsf = load_phylocsf("PhyloCSF_scores.tsv.gz")

# Merge datasets by transcript_id
import pandas as pd
df_merged = df_appris.merge(df_sequences, on="transcript_id") \
                     .merge(df_annotation, on="transcript_id") \
                     .merge(df_qpfam, on="transcript_id", how="left") \
                     .merge(df_phylocsf, on="transcript_id", how="left")