PLINK Genotype File Formats

What is PLINK and Why Do We Need It?

PLINK is a free, open-source toolset designed for genome-wide association studies (GWAS) and population genetics analysis.

Why PLINK Exists

When you're dealing with genotype data from thousands (or millions) of people across hundreds of thousands (or millions) of genetic variants, you face several problems:

1. File size: Raw genotype data is MASSIVE
2. Processing speed: Reading and analyzing this data needs to be fast
3. Standardization: Different labs and companies produce data in different formats
4. Analysis tools: You need efficient ways to compute allele frequencies, test for associations, filter variants, etc.

PLINK solves these problems by providing:

• Efficient binary file formats (compact storage)
• Fast algorithms for common genetic analyses
• Format conversion tools
• Quality control utilities

When You'd Use PLINK

• Analyzing data from genotyping chips (Illumina, Affymetrix)
• Running genome-wide association studies (GWAS)
• Computing population genetics statistics
• Quality control and filtering of genetic variants
• Converting between different genotype file formats

PLINK Binary Format (.bed/.bim/.fam)

This is PLINK's primary format - a set of three files that work together. It's called "binary" because the main genotype data is stored in a compressed binary format rather than human-readable text.

The .fam File (Family/Sample Information)

The .fam file contains information about each individual (sample) in your study. It has 6 columns with NO header row.

Format:

FamilyID  IndividualID  FatherID  MotherID  Sex  Phenotype

Example .fam file:

FAM001  IND001  0  0  1  2
FAM001  IND002  0  0  2  1
FAM002  IND003  IND004  IND005  1  -9
FAM002  IND004  0  0  1  1
FAM002  IND005  0  0  2  1

Column Breakdown:

Column 1: Family ID

• Groups individuals into families
• Can be the same as Individual ID if samples are unrelated
• Example: FAM001, FAM002

Column 2: Individual ID

• Unique identifier for each person
• Must be unique within each family
• Example: IND001, IND002

Column 3: Paternal ID (Father)

• Individual ID of the father
• 0 = father not in dataset (unknown or not genotyped)
• Used for constructing pedigrees and family-based analyses

Column 4: Maternal ID (Mother)

• Individual ID of the mother
• 0 = mother not in dataset
• Must match an Individual ID if the parent is in the study

Column 5: Sex

• 1 = Male
• 2 = Female
• 0 = Unknown sex
• Other codes (like -9) are sometimes used for unknown, but 0 is standard

Column 6: Phenotype

• The trait you're studying (disease status, quantitative trait, etc.)
• For binary (case-control) traits:
  • 1 = Control (unaffected)
  • 2 = Case (affected)
  • 0 or -9 = Missing phenotype
• For quantitative traits: Any numeric value
• -9 = Standard missing value code

Important Notes About Special Codes:

0 (Zero):

• In Parent columns: Parent not in dataset
• In Sex column: Unknown sex
• In Phenotype column: Missing phenotype (though -9 is more common)

-9 (Negative nine):

• Universal "missing data" code in PLINK
• Most commonly used for missing phenotype
• Sometimes used for unknown sex (though 0 is standard)

Why these codes matter:

• PLINK will skip individuals with missing phenotypes in association tests
• Parent information is crucial for family-based tests (like TDT)
• Sex information is needed for X-chromosome analysis

The .bim File (Variant Information)

The .bim file (binary marker information) describes each genetic variant. It has 6 columns with NO header row.

Format:

Chromosome  VariantID  GeneticDistance  Position  Allele1  Allele2

Example .bim file:

1   rs12345    0    752566    G    A
1   rs67890    0    798959    C    T
2   rs11111    0    1240532   A    G
3   rs22222    0    5820321   T    C
X   rs33333    0    2947392   G    A

Column Breakdown:

Column 1: Chromosome

• Chromosome number: 1-22 (autosomes)
• Sex chromosomes: X, Y, XY (pseudoautosomal), MT (mitochondrial)
• Example: 1, 2, X

Column 2: Variant ID

• Usually an rsID (reference SNP ID from dbSNP)
• Format: rs followed by numbers (e.g., rs12345)
• Can be any unique identifier if rsID isn't available
• Example: chr1:752566:G:A (chromosome:position:ref:alt format)

Column 3: Genetic Distance

• Position in centimorgans (cM)
• Measures recombination distance, not physical distance
• Often set to 0 if unknown (very common)
• Used in linkage analysis and some phasing algorithms

Column 4: Base-Pair Position

• Physical position on the chromosome
• Measured in base pairs from the start of the chromosome
• Example: 752566 means 752,566 bases from chromosome start
• Critical for genome builds: Make sure you know if it's GRCh37 (hg19) or GRCh38 (hg38)!

Column 5: Allele 1

• First allele (often the reference allele)
• Single letter: A, C, G, T
• Can also be I (insertion), D (deletion), or 0 (missing)

Column 6: Allele 2

• Second allele (often the alternate/effect allele)
• Same coding as Allele 1

Important Notes:

Allele coding:

• These alleles define what genotypes mean in the .bed file
• Genotype AA means homozygous for Allele1
• Genotype AB means heterozygous
• Genotype BB means homozygous for Allele2

Strand issues:

• Alleles should be on the forward strand
• Mixing strands between datasets causes major problems in meta-analysis
• Always check strand alignment when combining datasets!

The .bed File (Binary Genotype Data)

The .bed file contains the actual genotype calls in compressed binary format. This file is NOT human-readable - you can't open it in a text editor and make sense of it.

Key characteristics:

Why binary?

• Space efficiency: A text file with millions of genotypes is huge; binary format compresses this dramatically
• Speed: Computer can read binary data much faster than parsing text
• Example: A dataset with 1 million SNPs and 10,000 people:
  • Text format (.ped): ~30 GB
  • Binary format (.bed): ~2.4 GB

What's stored:

• Genotype calls for every individual at every variant
• Each genotype is encoded efficiently (2 bits per genotype)
• Encoding:
  • 00 = Homozygous for allele 1 (AA)
  • 01 = Missing genotype
  • 10 = Heterozygous (AB)
  • 11 = Homozygous for allele 2 (BB)

SNP-major vs. individual-major:

• PLINK binary files are stored in SNP-major mode by default
• This means genotypes are organized by variant (all individuals for SNP1, then all individuals for SNP2, etc.)
• More efficient for most analyses (which process one SNP at a time)

You never edit .bed files manually - always use PLINK commands to modify or convert them.

PLINK Text Format (.ped/.map)

This is the original PLINK format. It's human-readable but much larger and slower than binary format. Mostly used for small datasets or when you need to manually inspect/edit data.

The .map File (Variant Map)

Similar to .bim but with only 4 columns.

Format:

Chromosome  VariantID  GeneticDistance  Position

Example .map file:

1   rs12345    0    752566
1   rs67890    0    798959
2   rs11111    0    1240532
3   rs22222    0    5820321

Notice: NO allele information in .map files (unlike .bim files).

The .ped File (Pedigree + Genotypes)

Contains both sample information AND genotype data in one large text file.

Format:

FamilyID  IndividualID  FatherID  MotherID  Sex  Phenotype  [Genotypes...]

The first 6 columns are identical to the .fam file. After that, genotypes are listed as pairs of alleles (one pair per SNP).

Example .ped file:

FAM001  IND001  0  0  1  2  G G  C T  A G  T T
FAM001  IND002  0  0  2  1  G A  C C  A A  T C
FAM002  IND003  0  0  1  1  A A  T T  G G  C C

Genotype Encoding:

Each SNP is represented by two alleles separated by a space:

• G G = Homozygous for G allele
• G A = Heterozygous (one G, one A)
• A A = Homozygous for A allele
• 0 0 = Missing genotype

Important: The order of alleles in heterozygotes doesn't matter (G A = A G).

Problems with .ped format:

• HUGE files for large datasets (gigabytes to terabytes)
• Slow to process (text parsing is computationally expensive)
• No explicit allele definition (you have to infer which alleles exist from the data)

When to use .ped/.map:

• Small datasets (< 1,000 individuals, < 10,000 SNPs)
• When you need to manually edit genotypes
• Importing data from older software
• Best practice: Convert to binary format (.bed/.bim/.fam) immediately for analysis

Transposed Format (.tped/.tfam)

This format is a "transposed" version of .ped/.map. Instead of one row per individual, you have one row per SNP.

The .tfam File

Identical to .fam file - contains sample information.

Format:

FamilyID  IndividualID  FatherID  MotherID  Sex  Phenotype

The .tped File (Transposed Genotypes)

Each row represents one SNP, with genotypes for all individuals.

Format:

Chromosome  VariantID  GeneticDistance  Position  [Genotypes for all individuals...]

Example .tped file:

1  rs12345  0  752566  G G  G A  A A  G G  A A
1  rs67890  0  798959  C T  C C  T T  C T  C C
2  rs11111  0  1240532 A G  A A  G G  A G  A A

The first 4 columns are like the .map file. After that, genotypes are listed for all individuals (2 alleles per person, space-separated).

When to use .tped/.tfam:

• When your data is organized by SNP rather than by individual
• Converting from certain genotyping platforms
• Some imputation software prefers this format
• Still text format so same size/speed issues as .ped

Long Format

Long format (also called "additive" or "dosage" format) represents genotypes as numeric values instead of allele pairs.

Format options:

Additive coding (most common):

FamilyID  IndividualID  VariantID  Genotype
FAM001    IND001        rs12345    0
FAM001    IND001        rs67890    1
FAM001    IND001        rs11111    2
FAM001    IND002        rs12345    1

Numeric genotype values:

• 0 = Homozygous for reference allele (AA)
• 1 = Heterozygous (AB)
• 2 = Homozygous for alternate allele (BB)
• NA or -9 = Missing

Why long format?

• Easy to use in statistical software (R, Python pandas)
• Flexible for merging with other data (phenotypes, covariates)
• Good for database storage (one row per observation)
• Can include dosages for imputed data (values between 0-2, like 0.85)

Downsides:

• MASSIVE file size (one row per person per SNP)
• Example: 10,000 people × 1 million SNPs = 10 billion rows
• Not practical for genome-wide data without compression

When to use:

• Working with a small subset of SNPs in R/Python
• Merging genotypes with other tabular data
• Machine learning applications where you need a feature matrix

Variant Call Format (VCF)

VCF is the standard format for storing genetic variation from sequencing data. Unlike genotyping arrays (which only check specific SNPs), sequencing produces all variants, including rare and novel ones.

Key characteristics:

Comprehensive information:

• Genotypes for all samples at each variant
• Quality scores for each call
• Read depth, allele frequencies
• Functional annotations
• Multiple alternate alleles at the same position

File structure:

• Header lines start with ## (metadata about reference genome, samples, etc.)
• Column header line starts with #CHROM (defines columns)
• Data lines: One per variant

Standard VCF columns:

#CHROM  POS     ID         REF  ALT     QUAL  FILTER  INFO           FORMAT  [Sample genotypes...]
1       752566  rs12345    G    A       100   PASS    AF=0.23;DP=50  GT:DP   0/1:30  1/1:25  0/0:28

Column Breakdown:

CHROM: Chromosome (1-22, X, Y, MT)

POS: Position on chromosome (1-based coordinate)

ID: Variant identifier (rsID or . if none)

REF: Reference allele (what's in the reference genome)

ALT: Alternate allele(s) - can be multiple, comma-separated

• Example: A,T means two alternate alleles

QUAL: Quality score (higher = more confident call)

• Phred-scaled: QUAL=30 means 99.9% confidence
• . if unavailable

FILTER: Quality filter status

• PASS = passed all filters
• LowQual, HighMissing, etc. = failed specific filters
• . = no filtering applied

INFO: Semicolon-separated annotations

• AF=0.23 = Allele frequency 23%
• DP=50 = Total read depth
• AC=10 = Allele count
• Many possible fields (defined in header)

FORMAT: Describes the per-sample data fields

• GT = Genotype
• DP = Read depth for this sample
• GQ = Genotype quality
• Example: GT:DP:GQ

Sample columns: One column per individual

• Data corresponds to FORMAT field
• Example: 0/1:30:99 means heterozygous, 30 reads, quality 99

Genotype Encoding in VCF:

GT (Genotype) format:

• 0/0 = Homozygous reference (REF/REF)
• 0/1 = Heterozygous (REF/ALT)
• 1/1 = Homozygous alternate (ALT/ALT)
• ./. = Missing genotype
• 1/2 = Heterozygous with two different alternate alleles
• 0|1 = Phased genotype (pipe | instead of slash /)

Phased vs. unphased:

• / = unphased (don't know which allele came from which parent)
• | = phased (know parental origin)
• 0|1 means reference allele from parent 1, alternate from parent 2

Compressed VCF (.vcf.gz):

VCF files are usually gzipped and indexed:

• .vcf.gz = compressed VCF (much smaller)
• .vcf.gz.tbi = tabix index (allows fast random access)
• Tools like bcftools and vcftools work directly with compressed VCFs

Example sizes:

• Uncompressed VCF: 100 GB
• Compressed .vcf.gz: 10-15 GB
• Always work with compressed VCFs!

When to use VCF:

• Sequencing data (whole genome, exome, targeted)
• When you need detailed variant information
• Storing rare and novel variants
• Multi-sample studies with complex annotations
• NOT typical for genotyping array data (use PLINK binary instead)

Oxford Format (.gen / .bgen + .sample)

Developed by the Oxford statistics group, commonly used in UK Biobank and imputation software (IMPUTE2, SHAPEIT).

The .sample File

Contains sample information, similar to .fam but with a header row.

Format:

ID_1 ID_2 missing sex phenotype
0 0 0 D B
IND001 IND001 0 1 2
IND002 IND002 0 2 1

First two rows are special:

• Row 1: Column names
• Row 2: Data types
  • D = Discrete/categorical
  • C = Continuous
  • B = Binary
  • 0 = Not used

Subsequent rows: Sample data

• ID_1: Usually same as ID_2 for unrelated individuals
• ID_2: Sample identifier
• missing: Missingness rate (usually 0)
• sex: 1=male, 2=female
• phenotype: Your trait of interest

The .gen File (Genotype Probabilities)

Stores genotype probabilities rather than hard calls. This is crucial for imputed data where you're not certain of the exact genotype.

Format:

Chromosome  VariantID  Position  Allele1  Allele2  [Genotype probabilities for all samples...]

Example .gen file:

1  rs12345  752566  G  A  1 0 0  0.95 0.05 0  0 0.1 0.9

Genotype Probability Triplets:

For each sample, three probabilities (must sum to 1.0):

• P(AA) = Probability of homozygous for allele 1
• P(AB) = Probability of heterozygous
• P(BB) = Probability of homozygous for allele 2

Example interpretations:

• 1 0 0 = Definitely AA (100% certain)
• 0 0 1 = Definitely BB (100% certain)
• 0 1 0 = Definitely AB (100% certain)
• 0.9 0.1 0 = Probably AA, might be AB (uncertain genotype)
• 0.33 0.33 0.33 = Completely uncertain (missing data)

Why probabilities matter:

• Imputed genotypes aren't perfectly certain
• Better to use probabilities than picking "best guess" genotype
• Allows proper statistical modeling of uncertainty
• Example: If imputation says 90% chance of AA, 10% chance AB, you should account for that uncertainty

The .bgen File (Binary Gen)

Binary version of .gen format - compressed and indexed for fast access.

Key features:

• Much smaller than text .gen files
• Includes variant indexing for rapid queries
• Supports different compression levels
• Stores genotype probabilities (like .gen) or dosages
• Used by UK Biobank and other large biobanks

Associated files:

• .bgen = Main genotype file
• .bgen.bgi = Index file (for fast lookup)
• .sample = Sample information (same as with .gen)

When to use Oxford format:

• Working with imputed data
• UK Biobank analyses
• Using Oxford software (SNPTEST, QCTOOL, etc.)
• When you need to preserve genotype uncertainty

Converting to PLINK:

• PLINK2 can read .bgen files
• Can convert to hard calls (loses probability information)
• Or use dosages (keeps uncertainty as 0-2 continuous values)

23andMe Format

23andMe is a direct-to-consumer genetic testing company. Their raw data format is simple but NOT standardized for research use.

Format:

# rsid    chromosome    position    genotype
rs12345    1    752566    AG
rs67890    1    798959    CC
rs11111    2    1240532   --

Column Breakdown:

rsid: Variant identifier (rsID from dbSNP)

chromosome: Chromosome number (1-22, X, Y, MT)

• Note: Sometimes uses 23 for X, 24 for Y, 25 for XY, 26 for MT

position: Base-pair position

• Warning: Build version (GRCh37 vs GRCh38) is often unclear!
• Check the file header or 23andMe documentation

genotype: Two-letter allele call

• AG = Heterozygous
• AA = Homozygous
• -- = Missing/no call
• DD or II = Deletion or insertion (rare)

Important Limitations:

Not standardized:

• Different builds over time (some files are GRCh37, newer ones GRCh38)
• Allele orientation issues (forward vs. reverse strand)
• Variant filtering varies by chip version

Only genotyped SNPs:

• Typically 500k-1M SNPs (depending on chip version)
• No imputed data in raw download
• Focused on common variants (rare variants not included)

Missing quality information:

• No quality scores
• No read depth or confidence metrics
• "No call" (--) doesn't tell you why it failed

Privacy and consent issues:

• Users may not understand research implications
• IRB approval needed for research use
• Cannot assume informed consent for specific research

Converting 23andMe to PLINK:

Many online tools exist, but be careful:

1. Determine genome build (critical!)
2. Check strand orientation
3. Handle missing genotypes (-- → 0 0)
4. Verify chromosome coding (especially X/Y/MT)

Typical workflow:

# Convert to PLINK format (using a conversion script)
python 23andme_to_plink.py raw_data.txt

# Creates .ped and .map files
# Then convert to binary
plink --file raw_data --make-bed --out data

When you'd use 23andMe data:

• Personal genomics projects
• Ancestry analysis
• Polygenic risk score estimation
• Educational purposes
• NOT suitable for: Clinical decisions, serious GWAS (too small), research without proper consent

Summary: Choosing the Right Format

General recommendations:

1. For genotyping array data: Use PLINK binary format (.bed/.bim/.fam)
2. For sequencing data: Use compressed VCF (.vcf.gz)
3. For imputed data: Use Oxford .bgen or VCF with dosages
4. For statistical analysis: Convert subset to long format
5. For personal data: Convert 23andMe to PLINK, but carefully

File conversions:

• PLINK can convert between most formats
• Always document your conversions (genome build, strand, filters)
• Verify a few variants manually after conversion
• Keep original files - conversions can introduce errors