AIM Output Details

Output Files

Following the execution of AIM, you will receive four output files, which are saved to the designated output directory.

AIM output files

File Name	Explanation
*_default_predictions.csv	Prediction results from AIM Default mode.
*_recessive_predictions.csv	Prediction results from AIM Recessive mode. Each row is a pair of variants.
*_nd_predictions.csv	Prediction results from AIM Novel Disease Gene mode.
*_nd_recessive_predictions.csv	Prediction results from AIM Novel Disease Gene + Recessive mode.

Output Explanation

Each of the output files mentioned above, contains:

• Variant ID

• Annotation information (features)

• Prediction results: score, rank, and confidence

Prediction Results

predict

Prediction score from AIM

ranking

Maximum ranking of all variants by prediction scores

confidence

Confidence score. It’s the z-score of AIM score in our inner cohort.

confidence level

High / Medium / Low. We set confidence score cutoff based on precision and recall.

Variant Annotation / AIM Features

AIM takes information from various database and perform feature engineering, to make the final prediction. Here, we provide both raw and engineered features as part of the output.

In making the decision, AIM takes various types of information into consideration.

Conservation

GERPpp_RS

GERP++ RS score, the larger the score, the more conserved the site. Scores range from -12.3 to 6.17

LRT_Omega

Estimated nonsynonymous-to-synonymous-rate ratio (Omega, reported by LRT)

LRT_score

The original LRT two-sided p-value (LRTori), ranges from 0 to 1.

phyloP100way_vertebrate

phyloP (phylogenetic p-values) conservation score based on the multiple alignments of 100 vertebrate genomes (including human). The larger the score, the more conserved the site. Scores range from -20.0 to 10.003 in dbNSFP.

Constrain

hom

Number of homozygotes variant in gnomAD

decipherVarFound

0/1. Whether the variant is found in the a deletion of the DECIPHER control database.

dgvVarFound

0/1. Whether the variant is found in a deletion of the DGV database.

conservationScoreDGV

1/3 (Low/High). If DGV subtype is Loss or Deletion, score will be 1. Otherwise 3.

gnomadGeneOELof

observed/expected ratio of loss-of-function variants in gnomAD database.

gnomadGeneOELofUpper

The upper bound of the confidence interval for OE LoF.

conservationScoreOELof

1/2 (Low/High). If gnomadGeneOELofUpper < 0.35, score is 1; otherwise 2.

gnomadGenePLI

pLI score stands for the “probability of being loss-of-function intolerant” in gnomAD.

gnomadGeneZscore

The gene z-score in gnomAD is related to missense variants and reflects how many standard deviations the observed count of missense variants is from the expected count. This metric can help identify genes that are under selective pressure and may be related to diseases.

conservationScoreGnomad

1/2 (Low/High). If both gnomadAF and gnomadAFg are less than 0.01, score is high; otherwise low.

Disease Database

CLASS

CLASS from HGMD

DM: disease-causing mutation; DM? Likely disease-causing, but with questionable pathogenicity

clinVarGeneFound

0/1, whether or not variant gene is found in ClinVar.

clinVarVarFound

0/1, whether or not variant itself is found in ClinVar.

curationScoreClinVar

1/2/3 (Low/Medium/High), curated using ClinVar significance description.

isB/LB

0/1. Whether ClinVar significance description contains benign and no conflicting interpretation.

isP/LP

Float ranging 0-1. Among all descriptions in ClinVar about this variant, proportion of pathogenic ones.

clinvarNumB

Proportion of benign variants in the variant gene.

clinvarNumLB

Proportion of likely benign + benign variants in the variant gene.

clinvarNumLP

Proportion of likely pathogenic + pathogenic variants in the variant gene.

clinvarNumP

Proportion of pathogenic variants in the variant gene.

hgmdGeneFound

0/1, whether or not variant gene is found in HGMD.

hgmdVarFound

0/1, whether or not variant itself is found in HGMD.

curationScoreHGMD

1/2/3 (Low/Medium/High), curated with hgmdGeneFound and hgmdVarFound.

omimGeneFound

0/1, whether or not variant gene is found in OMIM.

omimVarFound

0/1, whether or not variant itself is found in OMIM.

curationScoreOMIM

1/2/3 (Low/Medium/High), curated with omimGeneFound and omimVarFound.

dominant

0/1. Whether the variant gene is annotated as dominant in OMIM.

recessive

0/1. Whether the variant gene is annotated as recessive in OMIM

hgmd_rs

HGMD rank score, interpreted as relative probabilities of pathogenicity.

c_ClinVar_*

Expansions of variant annotation from ClinVar. One-hot encoded.

c_CLNREVSTAT

The ClinVar Review status for the same protein change in ClinVar

c_HGMD_Exp_*

Expansions of variant annotation from HGMD. One-hot encoded.

c_isBLB

The original variant is annotated as Benign in ClinVar

c_isPLP

The original variant is annotated as Pathogenic or likely pathogenic in ClinVar

c_RANKSCORE

The HGMD RANKSCORE adapted from the original HGMD database

nc_ClinVar_Exp

Non-coding variant expansion (2bp upstream or downstream of the original variants position)

nc_CLNREVSTAT

non-coding variant expansion (2bp upstream or downstream of the original variants position)

nc_HGMD_Exp

non-coding variant expansion (2bp upstream or downstream of the original variants position)

nc_isBLB

The original variant is annotated as Benign in ClinVar

nc_isPLP

The original variant is annotated as Pathogenic or likely pathogenic in ClinVar

nc_RANKSCORE

Variant Impact

cons_*

Variant consequence type is one-hot encoded. Complete list:

‘transcript_ablation’, ‘splice_acceptor_variant’, ‘splice_donor_variant’, ‘stop_gained’, ‘frameshift_variant’, ‘stop_lost’, ‘start_lost’, ‘transcript_amplification’, ‘inframe_insertion’, ‘inframe_deletion’, ‘missense_variant’, ‘protein_altering_variant’, ‘splice_region_variant’, ‘splice_donor_5th_base_variant’, ‘splice_donor_region_variant’

IMPACT

Integer 0-4 (None, Modifier, Low, Moderate, High). Subjective impact classification of consequence type.

IMPACT.from.Tier

In Silico Prediction

CADD_phred

CADD Phred score

DANN_score

DANN score

fathmm_MKL_coding_score

fathmm-MKL coding socre from dbNSFP

FATHMM_score

FATHMM score from dbNSFP, minimum value selected.

M_CAP_score

M-CAP score

MutationAssessor_score

MutationAssessor score, maximum value selected.

Polyphen2_HDIV_score

Polyphen2 HDIV score, maximum value selected.

Polyphen2_HVAR_score

Polyphen2 HVAR score, maximum value selected.

REVEL_score

REVEL score, maximum value selected.

SIFT_score

SIFT score, minimum value selected.

Inferred Inheritance

No.Var.H

Gene level, Number of High IMPACT variants in the patient for candidate gene

No.Var.HM

Gene level, Number of High or Moderate IMPACT variants in the patient for candidate gene

No.Var.L

Gene level, Number of Low IMPACT variants in the patient for candidate gene

No.Var.M

Gene level, Number of Moderate IMPACT variants in the patient for candidate gene

TierAD

1~4, Dominant Inheritance Score. The lower the more pathogenic

TierAR

1~4, Recessive Inheritance Score. The lower the more pathogenic

TierAR.adj

1~4, Adjusted Recessive Inheritance Score. For a candidate gene, if a rare intronic variant observed together with a high IMPACT variant, adjusted

AD.matched

0/1, TierAD <= 2 and dominant == 1

AR.matched

0/1, TierAR <= 2 and recessive == 1

zyg

Variant zygosity, 1: heterozygous, 2: homozygous.

Minor Allele Frequency

ESP6500_AA_AF

ESP6500 African American Allele Frequency

ESP6500_EA_AF

ESP6500 European American Allele Frequency

gnomadAF

gnomAD exome Allele Frequency

gnomadAFg

gnomAD genome Allele Frequency

Phenotype Matching

clinVarSymMatchFlag

0/1, whether OMIM variant phenotype matches condition in ClinVar.

hgmdSymptomSimScore

Similarity score between patient phenotype and variant phenotype in HGMD.

hgmdSymMatchFlag

0/1, whether hgmdSymptomSimScore >= 0.2

omimSymptomSimScore

Similarity score between patient phenotype and variant phenotype in OMIM.

omimSymMatchFlag

0/1, whether omimSymptomSimScore >= 0.2

phrank

Phrank measures phenotype sets similarity of the patient phenotype with phenotype linked to a candidate gene.

diffuse_Phrank_STRING

A phenotype score is derived through network diffusion, utilizing the String network and employing the Phrank score as the initial seed score.

others

simple_repeat

0/1, whether variant is in simple repeat regions.

spliceAImax

Maximum of SpliceAI score among DS_AG, DS_AL, DS_DG, and DS_DL.