TRAINING & RESOURCES
See the workflow.
Understand each decision.
Choose a short introduction or the complete narrated module walkthrough, then go directly to the section you need.
Complete walkthrough: installation, homepage and four selected research workflows. English captions are included in the picture.
Read the complete narration transcript
Introduction
Behind every DNA profile is a human story—a question of identity, a family connection, or a search for answers. Probabilix was created to help scientists interpret that story with clarity, care and scientific discipline.
Developed by Probabilix Technologies Limited, Probabilix is a DNA interpretation platform designed to bring parentage testing, kinship analysis, forensic mixture interpretation and missing-person identification into one connected environment.
Its vision is to bridge the gap between the DNA data a laboratory generates and the answers people need. By combining statistical analysis with structured review and traceable reporting, Probabilix aims to help laboratories make complex genetic evidence easier to examine, understand and explain.
From investigating biological relationships to advancing forensic research, Probabilix represents an ambition to contribute to the future of DNA science.
Every profile deserves careful interpretation. Every answer deserves a sound scientific foundation.
Installation
This chapter gives guided Windows instructions, rather than a recording of Windows execution. Extract the entire release ZIP. Open INSTALL PROBABILIX WINDOWS dot bat, choose a new installation folder and install its isolated dependencies. The current package requires Python three point eleven or later with Tkinter; first dependency installation needs internet. Use the new shortcut or START PROBABILIX UNIFIED dot bat and leave the launcher console open. Use the browser address opened by that launcher. Do not combine individual files from different releases. If startup reports missing or changed files, extract the complete original package into another new folder.
Full Probabilix homepage
After installation, this is the full Probabilix homepage. All eight workspace areas are shown here, from the logo and navigation down to the footer. The homepage connects continuous intake and quality control, mixture research, mixture evidence, parentage and kinship, relationship research, missing-person research, validation and audit, and administration and support. In this video, we will explain Parentage and kinship first, followed by Mixture research, Missing-person research, and Relationship research.
Module 04 — Parentage and kinship | Parentage: prepare the example
Open Parentage and kinship comparisons. Leave Comparison on Duo Paternity. In Child, choose S Y N CHILD STUTTER dot x l s x. In Alleged father, choose S Y N FATHER STUTTER dot x l s x. Use the synthetic frequency workbook supplied with this course in Optional allele-count workbook. Its filename explicitly says it is not population data. It contains artificial equal counts solely to make this exercise reproducible. Confirm the sample roles and research scope, then click Compare markers. In real work, use the controlled, appropriate population dataset and retain its revision and provenance.
Module 04 — Parentage and kinship | Read the pending result first
The first comparison displays twenty-seven markers but withholds the research combined paternity index because D five S eight one eight requires review. Displayed marker count is not the same as usable marker count. Read the blocker sentence above the conclusion and find the flagged locus below it. Do not interpret Unable to interpret as an exclusion. It means no usable combined index is currently available. Do not remove the flag by changing the source workbook or inventing frequencies. This lesson resolves a particular stutter-only review by examining the exported typed status, sequences and read ratios.
Module 04 — Parentage and kinship | What the D5S818 candidates mean
Scroll to D five S eight one eight sequence and stutter review. The child summary is twelve, thirteen. Allele twelve has nine hundred forty reads and allele thirteen has one thousand. Their ratio is ninety-four percent, above the workbook's eleven percent stutter setting. Another child signal, allele eleven, is untyped and has seventy-four reads, about seven point eight seven percent of allele twelve. The father's untyped allele ten has forty reads against eight hundred for allele eleven, or five percent. These are invented teaching values. A low ratio alone does not prove stutter; inspect sequence geometry, the source calls and laboratory evidence.
Module 04 — Parentage and kinship | Enter the reviewer and confirm
Enter TRAINING REVIEWER in Reviewer name. Tick I confirm this is my decision. The current parentage disposition workflow requires your name and this explicit confirmation. A supporting note or laboratory-rule reference is optional. Your confirmation is an accountable statement of your assessment; it does not create missing sequence evidence. Make one decision at a time. After saving a decision, the software resets the confirmation checkbox. Wait for confirmed and saved and the index recheck before ticking it again for the next signal. Moving too quickly can leave the next action without a valid confirmation.
Module 04 — Parentage and kinship | Retain the supported typed allele
For the child's allele twelve candidate, choose Reject stutter, include signal. The words are important: you are rejecting the stutter interpretation and keeping that signal in the reviewed set. In this constructed case, twelve is a reported typed allele with strong exported support. Excluding it would conflict with the existing summary genotype and keep the current-genotype calculation pending. This is why Accept stutter is not a universal resolve button. Check the row number, role and typed status before choosing. Wait until this exact candidate is confirmed and saved, and allow the automatic recalculation to finish.
Module 04 — Parentage and kinship | Exclude the supported untyped candidates
Tick the confirmation again after the previous save completes. For child allele eleven, choose Accept stutter, exclude signal. Wait for the saved message and recheck. Tick confirmation once more, then make the same supported exclusion for the father's untyped allele ten. In this exercise those decisions leave the reported child and father genotypes intact. In a real case, only make these choices when the source evidence and your review support them. If the evidence is ambiguous, retain the review hold. Never exclude a typed allele simply because its number is one repeat below another allele.
Module 04 — Parentage and kinship | Verify the recalculated full CPI
With all three supported stutter decisions saved, the exercise recalculates the full research combined paternity index to ten to the power twenty-seven. This very large value is a consequence of the deliberately artificial profiles and equal-count frequency table. It is not a performance claim. Check that the full-panel index is available, that no other locus remains blocked, and that the printed report reflects the current decisions. Use Print, save research report as P D F for the report, and Download review record for the decision history. Keep the source workbooks and their controlled population reference with the review.
Module 04 — Parentage and kinship | Imbalance is a separate review
Now substitute S Y N FATHER IMBALANCE for the father report. The full research index is withheld at D three S thirteen fifty-eight. Resolve that locus on its own evidence; clearing D five S eight one eight cannot clear an unrelated imbalance flag. The current tool provides Lab-reviewed genotype and Confirm reviewed genotype for supported calls. Enter your reviewer name and confirmation, select the correct sample role, and enter the supported two-copy genotype. A successful supported confirmation is saved and applied to the comparison. If the exported evidence cannot support that genotype, the calculation should remain pending. The course does not manufacture a resolution for missing evidence.
Module 04 — Parentage and kinship | Missing calls, frequencies and reduced panels
A missing call or missing population frequency is different from a stutter decision. Check the original export for a genuine resolved call, or the approved population source for the actual allele. Your laboratory rule is to omit a marker whose frequency is absent. The reduced-panel control can display a research index with eligible quality, missing-call or missing-frequency loci omitted. Always list what was omitted and why. This is a reduced-panel result; it does not resolve the full-panel result. An apparent biological incompatibility must be investigated as such, including sample identity, contamination and model assumptions. Do not conceal it by repeatedly removing inconvenient loci.
Module 04 — Parentage and kinship | The relationship comparison menu
The Comparison menu also includes Duo Maternity, Full siblings, Half siblings, Aunt or uncle and niece or nephew, and Grandparent and grandchild. Choose the documented relationship before uploading. For the two-person kinship modes the role labels change to Person one and Person two. A different hypothesis answers a different question. Do not choose whichever relationship gives the largest index. Retain the selected relationship and the unrelated alternative in the record. The current research calculations assume independent loci, no mutation and no population substructure adjustment. Those limitations remain even when every marker is technically usable.
Module 04 — Parentage and kinship | Worked full-sibling example
Select Full siblings and upload S Y N CHILD CLEAR as Person one and S Y N FATHER CLEAR as Person two. The filenames are just practice labels; selecting this mode assigns their comparison roles. Load the artificial frequency workbook, acknowledge the scope and click Compare markers. This constructed example produces a research kinship index of approximately two point seven eight times ten to the nineteenth. The selected prior is fifty percent. Relationship supported means the findings favour that selected relationship over the stated unrelated alternative under the model. It is not a validated laboratory classification or a confirmed real relationship.
Module 04 — Parentage and kinship | Practise the other two-person relationships
Using the same synthetic practice profiles, run Half siblings, then the avuncular option, then Grandparent and grandchild. Each change clears the previous displayed result; click Compare markers again. Confirm the role labels and included marker panel on each run. Compare the conclusions and assumptions rather than treating the values as interchangeable. For Duo Maternity, use the child and mother CLEAR reports in their documented roles. No biological relationship is established by this exercise. Its purpose is to teach the menu, inputs, marker review and correct interpretation of a relationship-specific research index.
Module 04 — Parentage and kinship | Trio paternity: current limit
Select Trio Paternity and add S Y N MOTHER CLEAR to Mother. Check the child and alleged father selections before comparing again. The current trio screen supports marker review, but it does not provide a numeric trio relationship index. The recorded screen states No numeric relationship index is available for this comparison. That message is a software capability limit. It is not fixed by confirming a disposition or adding a frequency workbook. Review maternal and paternal inheritance at each locus, retain any technical holds, and use a separately supported, appropriately validated model if a numeric trio statistic is required.
Module 04 — Parentage and kinship | Save exactly what was reviewed
Before printing any parentage or kinship research report, recheck the relationship, sample roles, marker panel, frequency revision and remaining flags. Changing inputs requires a new comparison. The result should distinguish the full panel from a reduced panel and preserve the original source flags even after a supported decision. Your name and confirmation document the disposition; the exported review history documents how it was reached. Keep the report labelled research only. A large index, a high displayed probability, or a saved decision does not by itself establish scientific validation or authorize a forensic casework report.
Module 02 — Mixture research | Estimate contributor count
Open Estimate contributor count and compare two versus three. Select S Y N MIXTURE two P. Screening starts automatically. This example returns a provisional two-contributor estimate, with twenty of twenty-seven autosomal loci meeting screening requirements. Read the explanation and threshold-sensitivity display. The count is a screening estimate, not proof of the true number of people. Low-level contributors may be missed, while artefacts can increase the apparent count. A count chosen for a subsequent analysis is an assumption that needs review against the profile and experimental context. Keep the contributor count consistent with both L R propositions.
Module 02 — Mixture research | Review count instability
Use Review individual loci to identify unusually high allele counts or insufficient support. Use Read-threshold sensitivity to see whether the suggested band changes as the screen changes its read cutoff. Compare two versus three contributors where that control is available. A stable band is a useful diagnostic, but it does not detect every hidden contributor. If the selected count conflicts with the screen, the deconvolution or L R module may withhold calculation. Investigate low-level alleles, imbalance, noise and sample quality. Do not repeatedly change the count just to produce a larger L R.
Module 02 — Mixture research | Reconstruct possible genotype groups
Open Deconvolve a D N A mixture. Choose the same synthetic mixture, select Two under Assumed contributors, confirm the mixture, count and research scope, then click Show genotype combinations. Leave the special archived-study options unselected; this course does not use O N C O M I X. The result analyses twenty loci and lists seven omissions. At each marker, the table displays possible unordered groups of contributor genotypes. It does not identify named people or automatically join a particular person's genotype across different loci. Keep that distinction clear when explaining deconvolution to a colleague.
Module 02 — Mixture research | Read uncertainty in deconvolution
At the first displayed marker, the leading synthetic group has about fifty-six point five nine percent model probability. The next group has about twenty-eight point two six percent. About five point six one percent belongs to groups outside the displayed top five. Read ties and omitted markers as well as the first row. Use DNA marker to change the locus and Model setting to compare the two assumptions. These probabilities are conditional on the selected count, exported reads and finite working prior. The Unseen category collects sequences outside the named bins; it does not estimate biological dropout or identify a missing allele.
Module 02 — Mixture research | Known-reference mixture trial
Open Run a known-reference mixture trial. Select the synthetic mixture. Under Known contributors, select S Y N REFERENCE and S Y N REFERENCE B together. Under Comparison references, choose S Y N NONCONTRIBUTOR. Confirm the roles and click Run research trial. The operator supplies the known group; the software does not discover ground truth. This practice run compares three groups across twenty shared reference loci. The known group has sixty matches and unique rank one under both exact sequence and allele number. Those are descriptive match rankings, not likelihood ratios or probabilities of contribution.
Module 02 — Mixture research | Marker-removal sensitivity and batch review
The trial report repeats its ranking after omitting each marker. This tests whether the descriptive ranking depends on a particular locus. In our simple constructed example, the known group remains unique rank one. Click Download result dot J S O N. Open Compare saved trial results and select the downloaded synthetic trial file. To compare several trials, reselect all their result files together. Lower rank is better in this descriptive comparison; at equal rank, fewer tied groups is better. This example reports the same rank and tie count for sequence and allele number, so it does not demonstrate a sequence advantage.
Module 02 — Mixture research | Live LR: state both propositions
Open Calculate a live research L R. Select the synthetic mixture and selected reference. Choose two contributors under both propositions. The screen defines H p as the selected person plus one unknown unrelated contributor, and H d as two unknown unrelated contributors. Confirm those are the explanations being compared. The same count applies to both. A reference matching some alleles is not by itself an L R calculation. The model must calculate the evidence probability under both explanations and state its population, dropout, background and other limits. This current implementation is an exploratory finite research model.
Module 02 — Mixture research | A restricted-panel hold and its action
The first run with the reduced-panel checkbox left empty withholds calculation. The message says that the omitted loci require review and a restricted-panel calculation needs explicit selection. Twenty loci are supported and seven are omitted in our synthetic fixture. Expand Review marker panel and omissions and examine the reasons. Only if a reduced research panel is appropriate, tick If necessary, calculate on at least fifteen supported loci, Show every omitted locus, then click Calculate research L R again. This is not permission to ignore failed controls or inadequate data. Fewer than fifteen supported loci remains a separate blocking condition.
Module 02 — Mixture research | Read every LR assumption row
The supported synthetic calculation shows about four point one five five times ten to the fifth under Baseline, nine point seven eight two times ten to the fifth under Higher stutter and background, and two point seven four five times ten to the sixth under Unobserved allele stress. Review all rows, panel omissions and model limits. An L R compares explanations; it is not the probability that the person contributed. If different assumptions place some ratios above one and others below one, the direction is assumption-sensitive and must be reported as such. Download the research calculation J S O N to preserve the full result.
Module 06 — Missing-person research | Register a missing-person research case
Open Missing-person case workflow. Click Load a synthetic example. This fills a fictitious unidentified profile, a matching direct reference and two biological parents across three markers, together with artificial frequencies. In real research, use a non-identifying case I D, specimen I Ds and custody references. Add each reference once, select the documented relationship, and review sample quality before setting Reviewed for research screening. A direct reference should actually come from the missing person. A relative is assigned the correct family role. Incorrect roles can make a mathematically correct comparison answer the wrong biological question.
Module 06 — Missing-person research | Direct and family comparisons are different
Confirm specimen roles, source records, relationships and scope, then click Run research comparisons. The synthetic direct comparison has three exact matches, zero discordant markers and a research L R of five hundred twelve. The family comparison has a research L R of sixty-four. These compare different explanations. The direct comparison asks same source versus an unrelated source. The family comparison asks whether the unidentified person occupies the specified family position versus an unrelated source. Do not multiply those two displayed results together as if independent. These three-marker teaching results are not adequate evidence for identification.
Module 06 — Missing-person research | QC holds and discordance
Change the unidentified sample's Q C status to Pending or held and run again. Both comparisons now show Sample Q C hold. The correct resolution is to complete the actual laboratory review, not merely select a more favourable menu value. Missing population frequencies prevent a numeric research comparison; they do not justify inventing a default population. Discordant direct calls need source and sample review. Family incompatibility requires checking roles, genotypes and model assumptions. Use the technical hold or request-repeat disposition when the cause remains unresolved. More relatives help only when their biological relationships and evidence quality are correctly documented.
Module 06 — Missing-person research | Record and export the missing-person review
After reviewing results, enter Reviewer I D and choose Disposition. The options include Awaiting review, Keep on technical hold, Request repeat or additional evidence, and Reviewed for research. Add concise notes that explain any remaining uncertainty, then click Download research case J S O N. That export preserves the local research result and review status. It is not an identification certificate. If you change input profiles, roles, frequencies or Q C, run the comparison again and review the new result. A previous disposition should not be assumed to approve changed inputs.
Module 05 — Relationship research | Relationship research workbench
Relationship research opens an older workbench containing genotype intake, mixture allele overlap and missing-person concordance diagnostics. It also exposes typed-read, non-typed sequence, motif and alignment review tools. Use the labelled action matching your question. Overlapping alleles are not evidence that a named individual contributed, and exact concordance is not an identification decision. For two-report reproducibility, verify that they truly represent the same donor and document independent runs or preparations. The software cannot authenticate those experimental facts from filenames. Keep claimed metadata and directly observed evidence separate in the record.
Module 05 — Relationship research | Synthetic extended pedigree example
Within the workbench, Open legacy kinship research workbench exposes five diagnostic models. Select synthetic full pedigree. The built-in example uses illustrative A and B alleles, equal frequencies and a synthetic mutation matrix. Acknowledge research use, then click Run research diagnostics. The demonstrated conditional probability is one quarter for the stated untyped-full-sibling-parent hypothesis. The output deliberately leaves combined ratio, L R and parentage probability unavailable. This is a teaching diagnostic, not a multi-locus case conclusion. The separate Open extended pedigree research button launches a Windows tool and is covered as a guided hand-off, not a demonstrated Windows execution.
COMMON REVIEW QUESTIONS
Useful answers.
A clear next step.
Why can a full research CPI remain pending?
Read the named blocker first. A confirmed disposition cannot create an absent call, a missing population frequency or a numeric model outside the screen’s supported scope.
Which stutter action retains a signal?
“Reject stutter · include signal” retains the reviewed signal. “Accept stutter · exclude signal” removes it. The source evidence, role and typed status determine the supported choice.
What does “Accept and implement” change?
In continuous intake, it changes the stated research preview threshold. In LR comparison, it changes the preferred existing research assumption row. Those actions do not automatically retrain the forensic LR engine.
Does a large LR establish identity?
An LR compares the stated explanations under a model. It is not itself a probability of identity or an identification decision.
How should I request a demonstration?
Describe your organization, platform and intended use through the contact form. It prepares an email draft to the published contact, with a copy and download option.
LET’S TALK SCIENCE
Bring your laboratory’s questions.
Explore the workflow, its assumptions and the evidence relevant to your intended use.
