Automated STR Data Analysis:Validation Studies

Automated AnalysisDatabasing ValidationCasework Studies

Mark W. Perlin (Cybergenetics, Pittsburgh, PA)David Coffman (Florida Department of Law Enforcement, Tallahassee, FL)

Cecelia A. Crouse & Felipe Konotop (Palm Beach County Sheriff’s Office, FL) Jeffrey D. Ban (Division of Forensic Science, Richmond, VA)

NIJ grants 2000-IJ-CX-K005 & 2001-IJ-CX-K003

Reviewing STR DataReviewing STR DataHuman data review bottleneckHuman data review bottleneck

Computer AutomationComputer AutomationQuality AssuranceQuality AssuranceDatabase IntegrityDatabase IntegrityCasework & MixturesCasework & Mixtures

Key goals:Key goals:• • no errorno error• • high throughputhigh throughput• • small staffsmall staff

TrueAllele™ TechnologyTrueAllele™ TechnologyEliminates STR human review bottleneckEliminates STR human review bottleneck

Gel-based, orGel-based, orSequencer, orSequencer, orCapillaryCapillary

Raw STRRaw STRData Data INPUTINPUT

QualityQualityAssuredAssuredProfilesProfiles

DatabaseDatabaseOUTPUTOUTPUT

Fully AutomatedFully Automated(on Mac/PC/Unix)(on Mac/PC/Unix)Color SeparationColor SeparationImage ProcessingImage ProcessingLane TrackingLane TrackingSignal AnalysisSignal AnalysisLadder BuildingLadder BuildingPeak QuantificationPeak QuantificationAllele DesignationAllele DesignationQuality CheckingQuality CheckingCODIS ReportingCODIS Reporting

Protected by US patents 5,541,067 & 5,580,728 & 5,876,933 & 6,054,268Protected by US patents 5,541,067 & 5,580,728 & 5,876,933 & 6,054,268

Automated ProcessingAutomated Processing1.Input1.Input 2.Gel/CE2.Gel/CE 3.Allele3.Allele 4.Output4.Output

Quality AssuranceQuality Assurance

Rule SystemRule System

Good data. “Low” optical density?

User sets criteria.No need to review.No rules fired.

ABI/3100 plate MegaBACE

Multi-Platform EngineMulti-Platform Engine

Validation MethodsValidation Methods

1. Obtain original data1. Obtain original data2. Process data in TrueAllele ES2. Process data in TrueAllele ES (auto-setup, process run, Q/A, (auto-setup, process run, Q/A, call alleles, apply rules, check)call alleles, apply rules, check) computer: accept/reject/editcomputer: accept/reject/edit3. Review all data3. Review all data one person, many computersone person, many computers human: accept/reject/edithuman: accept/reject/edit4. Generate results & stats4. Generate results & stats

ExtractExtract

AmplifyAmplify

SeparateSeparate

OtherOther

Rule SettingsRule Settings

Hitachi + PowerPlexHitachi + PowerPlexHitachi FM/Bio2 & Promega PowerPlex 1.2Hitachi FM/Bio2 & Promega PowerPlex 1.2

Computer: ~75%* data, no review needed Computer: ~75%* data, no review needed Human: All these designations correctHuman: All these designations correct

~8,000 PBSO genotypes reviewed~8,000 PBSO genotypes reviewedTrueAllele performed all gel & allele processingTrueAllele performed all gel & allele processing

TrueAllele expert system can eliminateTrueAllele expert system can eliminatemost human review of gel STR datamost human review of gel STR data

Hitachi ResultsHitachi Results

Accept Edit RejectAccept Edit RejectR

ejec

t

E

dit

A

ccep

tR

ejec

t

E

dit

A

ccep

tHuman ReviewHuman Review

Co

mp

ute

r P

roce

ssC

om

pu

ter

Pro

cess

non-automation datanon-automation data

ABI/310 + Pro/CoFilerABI/310 + Pro/CoFilerABI 310 & ProfilerPlus/CofilerABI 310 & ProfilerPlus/Cofiler

Computer: ~85% data, no review needed Computer: ~85% data, no review needed Human: All proper designations correctHuman: All proper designations correct

~24,000 FDLE genotypes reviewed~24,000 FDLE genotypes reviewedTrueAllele performed all CE & allele processingTrueAllele performed all CE & allele processing

TrueAllele expert system can eliminateTrueAllele expert system can eliminatemost human review of CE STR datamost human review of CE STR data

310 Results310 Results

Accept Edit RejectAccept Edit RejectR

ejec

t

E

dit

A

ccep

tR

ejec

t

E

dit

A

ccep

tHuman ReviewHuman Review

Co

mp

ute

r P

roce

ssC

om

pu

ter

Pro

cess

94 genos/min94 genos/min

ABI/3700 + Pro/CoFilerABI/3700 + Pro/CoFilerABI 3700 & ProfilerPlus/CofilerABI 3700 & ProfilerPlus/Cofiler

Computer: ~85% data, no review needed Computer: ~85% data, no review needed Human: All TA/ES designations correctHuman: All TA/ES designations correct

~17,000 FDLE genotypes reviewed~17,000 FDLE genotypes reviewedTrueAllele performed all CE & allele processingTrueAllele performed all CE & allele processing

TrueAllele expert system can eliminateTrueAllele expert system can eliminatemost human review of CE-array STR datamost human review of CE-array STR data

3700 Results3700 Results

Accept Edit RejectAccept Edit RejectR

ejec

t

E

dit

A

ccep

tR

ejec

t

E

dit

A

ccep

tHuman ReviewHuman Review

Co

mp

ute

r P

roce

ssC

om

pu

ter

Pro

cess

76 genos/min76 genos/min

The UK FSS ExperienceThe UK FSS ExperienceGenerate STR DataGenerate STR Data

UK National DNA DatabaseUK National DNA Database

Person reviewsPerson reviewsa fractiona fraction

of the dataof the data

TrueAllele expert systemTrueAllele expert systemscores all STR data and scores all STR data and assesses data qualityassesses data quality

FSS ABI/377 ValidationFSS ABI/377 ValidationResourcesResources• • Data: 22,000 genotypes (SGMplus)Data: 22,000 genotypes (SGMplus)• • People: 6 reviewers + 6 managers People: 6 reviewers + 6 managers • • Time: 8 weeks work + 4 weeks reportTime: 8 weeks work + 4 weeks report

ComponentsComponents• • Peak height correlation (GS vs TA)Peak height correlation (GS vs TA)• • Establish baseline height (error-free)Establish baseline height (error-free)• • Designation accuracy (human vs TA)Designation accuracy (human vs TA)• • Network/computer environmentNetwork/computer environment• • QMS documentationQMS documentation

ResultsResults• • Greater yield with TAGreater yield with TA• • No errors on quality dataNo errors on quality data

Casework StudiesCasework Studies

NonmixtureNonmixtureMixtureMixtureRape KitsRape KitsDisastersDisastersLCN, SNPsLCN, SNPs

M.W. Perlin and B. Szabady, M.W. Perlin and B. Szabady, ““Linear mixture analysis: Linear mixture analysis: a mathematical approach to resolving mixed DNA samples,a mathematical approach to resolving mixed DNA samples,””

Journal of Forensic SciencesJournal of Forensic Sciences, November, 2001., November, 2001.

Statistical InformationStatistical Information

Degenerate SGM+ alleles (6x6x6x10x...x6)Degenerate SGM+ alleles (6x6x6x10x...x6) 100,000,000 feasible profiles100,000,000 feasible profilesCompute unknown minor contrib profilesCompute unknown minor contrib profiles @30% @30% 1 feasible profile1 feasible profile @10% @10% 100 feasible profiles100 feasible profiles

LMA increases identification power LMA increases identification power a million-fold; a million-fold; CODIS matchCODIS match

Prob{Prob{STR profile STR profile }}& peak quants& peak quants}}

Validation Data SetsValidation Data Sets

Collaborators:Collaborators: Florida, Virginia, Florida, Virginia, New York, FBI, UK, Private LabsNew York, FBI, UK, Private Labs

Data:Data: synthetic lab mixtures, synthetic lab mixtures, casework, rape kits, disasterscasework, rape kits, disasters

Studies:Studies: comparison, concordance, comparison, concordance,automated lab processesautomated lab processes

Input:Input: TrueAllele quantitated peaks TrueAllele quantitated peaks

Lab: 70% victim, Lab: 70% victim, 30% unknown30% unknownComputer: 71%, 29%Computer: 71%, 29%

Rape Kit: Unknown SuspectRape Kit: Unknown Suspect

Known victimKnown victimUnknown suspectUnknown suspectFlorida data (FDLE, PBSO)Florida data (FDLE, PBSO)9:1, 7:3, 5:5, 3:7, 1:9 ratios9:1, 7:3, 5:5, 3:7, 1:9 ratiosSix pairs of samplesSix pairs of samples

11//1010 22//66

Other Mixing ProportionsOther Mixing Proportions

Lab: 90% victim, 10% unknownComputer: 90%, 10%

Lab: 50% victim, 50% unknownComputer: 52%, 48% Lab & review automation

Disaster Data ReviewDisaster Data Review

ConclusionsConclusions

• • TrueAllele TrueAllele databasing validationdatabasing validation• • Reduce time, error, staff & costReduce time, error, staff & cost

• • Ongoing Ongoing casework validationcasework validation• • Automate: data review & lab workAutomate: data review & lab work• • Serve: police, courts, societyServe: police, courts, society• • Objective, comprehensiveObjective, comprehensive