EVAL_PREDICTIONS(1)

✨ eval_predictions — the vibe check your gene predictions have been BEGGING for ✨

okay bestie, so eval_predictions is a User Command, and it's literally the moment of truth for your gene-finding era. 💅

🌸 NAME 🌸

eval_predictions — compares predicted genes with "real" (or annotated) genes. no cap, that's the whole job and she does it flawlessly.

🎀 SYNOPSIS 🎀

eval_predictions -r <real_fname_list> -p <pred_fname_list> -l <seq_len_list> [OPTIONS]

💖 DESCRIPTION 💖

So here's the tea: eval_predictions takes your predicted genes and holds them up next to the "real" (or annotated) genes, and then it reports all the standard measures of prediction quality. It's giving ✨report card✨ energy.

The measures it spills, in order:

• nucleotide sensitivity (Sn)
• nucleotide specificity (Sp)
• approximate correlation (AC)
• correlation coefficient (CC)
• exon sensitivity (ESn)
• exon specificity (ESp)
• proportion of real exons correctly predicted (CRa)
• proportion of real exons partially predicted (PCa)
• proportion of real exons with overlapping predictions (OLa)
• missed exons (ME)
• proportion of predicted exons that are correct (CRp)
• proportion of predicted exons that are partially correct (PCp)
• proportion of predicted exons that overlap real ones (OLp)
• wrong exons (WE)

Every single one of these quantities is computed exactly as described in "Evaluation of Gene-Finding Programs on Mammalian Sequences," by Rogic et al. (Genome Research 11:817-832). That's the source material, periodt. 📚

And because math is math and the universe demands balance:

• CRa + PCa + OLa + ME = 1
• CRp + PCp + OLp + WE = 1

Also bestie, note this: each set (predicted and real) should consist of non-overlapping groups of features (see refeature). overlapping groups are not it. 🙈

🔥 OPTIONS 🔥

• -r <real_fname_list> — (required) the list of names of files defining real genes (GFF). this is the "truth" side of the equation, the blueprint, the moment. you literally cannot skip this one. 👑
• -p <pred_fname_list> — (required) the list of names of files defining predicted genes (GFF). must correspond in order to <real_fname_list>. so like, file 1 pairs with file 1, file 2 with file 2 — do NOT scramble the order, that's an L waiting to happen. 💀
• -l <seq_len_list> — (required) the list of lengths of sequences. needed to compute certain nucleotide-level statistics. no lengths, no nucleotide stats, sorry babe that's just how it is. 🧚
• -f <feat_list> — the list of names of all features denoting exon regions. by default, it's equal to the single name "CDS". if your exons go by another name, this is where you say so. 🎀
• -d <fname_prefix> — dumps the full coords of correct, partially correct, wrong, missed, and overlapping exons to a set of files having the specified file name prefix. it's the receipts, bestie — every exon category gets its own file. 🔍
• -n <nbases> — also reports stats on "nearly correct" exons, that is, incorrect exons whose boundaries are within <nbases> of being correct. the columns get labeled "NCa" and "NCp". it's giving "so close, so nearly iconic" energy. 😭
• -h — prints this help message. the classic. the main character of flags. whenever you're lost, she's there. 🫶

⚠️ NOTE: be sure stop codons are included in CDSs in both the predicted and real sets, or in neither set. mismatch here = chaos, and not the cute kind. 🌈

💜 a lil sign-off 💜

This page is eval_predictions version 1.4, dated May 2016, and honestly? still slaying. go forth, run your comparisons, get those sensitivities and specificities, and remember — the numbers don't lie but they DO need non-overlapping features. mwah. 💅✨