Technology

The POLEX™ platform

Proposed chemistry · Photon Oxidation

From raw ionic signal to a clinical-grade break map. We propose to use our novel photon-oxidation chemistry to mark every DNA break; nanopore sequencing and purpose-built AI resolve them at base resolution — single- and double-strand.

Live · motion

The break happens in milliseconds.
POLEX catches it.

Every break leaves a unique ionic signature as it passes through the pore. Our models learn the signatures and call breaks in real time. Base resolution. Native DNA. No amplification.

Platform · Pipeline

From pore to break map.

Our end-to-end pipeline takes raw nanopore signal and — through a stack of deep-learning models — produces base-resolution maps of DNA damage.

STAGE 01

Sample Prep

High-MW DNA extraction, optional enzymatic labeling of break ends.

STAGE 02

Nanopore Seq

Long reads on MinION / PromethION; raw ionic current captured live.

STAGE 03

Basecalling

Transformer-based basecaller translates squiggles into sequence + quality.

AI · Deep Learning
STAGE 04

Break Calling

Proprietary CNN + signal-level classifier detects SSBs and DSBs at base resolution.

AI · In2Cell Core
STAGE 05

Genome Mapping

Alignment, duplicate removal, context annotation across reference genomes.

STAGE 06

Interactive Report

Break landscapes, motif context, hotspot calls — exported as HTML / PDF.

Stage 01 · Sample prep

High-molecular-weight DNA extraction

Long reads need long DNA. We optimize lysis and bead cleanup to preserve fragments > 50 kb, with optional enzymatic labelling to flag damaged ends before sequencing.

Inputs

Tissue (fresh / FFPE), cell pellets, primary cells, organoids

Output

Quality-checked DNA library, 312 ng target

Stage 02 · Nanopore sequencing

Native-strand reads, no amplification

DNA passes through a protein nanopore. Ionic current disturbances encode the sequence. Subtler signals reveal breaks, modifications, and damage.

Hardware

Oxford Nanopore MinION / PromethION

Throughput

~1.2M reads per flow cell, streaming

Stage 03 · NeuroBase

AI basecaller specialised for damage

Our transformer basecaller is fine-tuned on break-containing reads, lifting recall around damage sites by ~30% vs stock Guppy / Dorado while maintaining Q scores.

Model

Transformer, 120M parameters

Training set

12M break-labelled reads

Stage 04 · POLEX Core

Proprietary break simulation

The signal-level CNN reads raw squiggle in parallel with basecalls, classifying every position as SSB, DSB, complex lesion, or intact. Per-read confidence scores feed downstream.

Sensitivity

> 92% recall on DSB calls

Specificity

> 98%, per-base

Stage 05 · Genome mapping

Context-aware alignment

Break calls are aligned to reference, annotated with genomic context (gene, exon, repeat, chromatin state, replication timing) via our ContextNet graph model.

References

GRCh38, T2T-CHM13, mouse mm39

Speed

~8 min per 30x genome

Stage 06 · Interactive report

Clinical-grade output

Break landscapes, hotspot calls, motif contexts, and repair-outcome signatures. Delivered as an interactive HTML or PDF that non-specialists can action.

Format

HTML, PDF, VCF, BED

Turnaround

~24 hours, sample-to-report

LIVE · VISUALIZATION Continuous zoom from cell to single base pair with detected breaks twinkling CELL · NUCLEUS · CHROMATIN CHROMOSOME · G-BANDED DOUBLE HELIX · MOLECULAR NETWORK SSB · SINGLE-STRAND BREAK A T G C A T T A C G G T A gap on one strand DSB · DOUBLE-STRAND BREAK A T G C A T T A C G T A gap on both strands BREAK DETECTION · BASE RESOLUTION CELL → CHROMOSOME → HELIX → BREAK
POLEX · Proprietary Diagnostic Engine

Signal to break call, end-to-end.

POLEX proprietary network — input features through hidden layers into the Core, out to break classification i_raw t_dwell k_5mer strand q_score context POLEX CORE SSB call DSB call motif / context INPUT HIDDEN 1 HIDDEN 2 ANALOG CORE OUTPUT BREAK MAP

From sample intake through basecalling and the POLEX Core simulation, each read is classified as SSB, DSB or complex, correlated against the tissue reference database, and delivered as an evidence-based clinical recommendation.

AI MODEL · 01

NeuroBase

Transformer basecaller fine-tuned on break-containing reads, higher recall around damage sites.

AI MODEL · 02

BreakSeer

Signal-level CNN reading raw squiggle alongside basecalls to flag SSBs and DSBs with per-read confidence.

AI MODEL · 03

ContextNet

Graph model placing each break in its genomic neighbourhood: motifs, chromatin, replication timing.

Stay ahead of DNA damage research

Monthly updates on the POLEX platform, publications, and open roles. No spam.