Multiplexed Capture Panels for Diagnostics and NGS

Team Dynamic Matrices | 15 February 2026

A single-target pull-down is a solved problem. Scaling it to fifty targets, or five hundred, is not, because the probes start interacting with each other and with the background, and the failures are per-target rather than global. A panel can report excellent pooled recovery while three of its targets contribute nothing. We design and manufacture catcher libraries for exactly this kind of panel, so here is what we have learned about making them work.

The four things that go wrong at scale

1. Cross-hybridisation. Any two probes that share a stretch of complementarity, or that share adapter or handle sequences, can bind each other. Probes with GC-rich regions bind partially to unintended targets. Both consume probe and generate off-target capture. With N probes there are N² pairwise interactions to worry about.

2. Uneven capture efficiency. Each probe has its own melting temperature, secondary structure and binding kinetics. One hybridisation temperature and time cannot be optimal for all of them. Probes with low Tm under-capture; probes with high GC over-capture and cross-react. In exome capture this shows up as the well-known GC bias in coverage (Bainbridge et al., Genome Biology, 2010).

3. Background that scales with capacity. Bead-based capture adds beads to add capacity. More bead surface means more non-specific adsorption of off-target nucleic acid, so signal-to-noise falls as the panel grows.

4. Redesign cost. Commercial bait sets are synthesised as a fixed pool. Adding a target, dropping one, or fixing a poorly performing probe means resynthesising the pool. Panels stop evolving because iteration is expensive.

Design principles that hold up

Match thermodynamics. Design all probes to fall within a narrow Tm window (we aim for ±3 °C) by adjusting length and, where needed, position. Tools such as Primer3, NUPACK and OligoAnalyzer handle this; the key is to compute Tm under the actual capture buffer conditions, not defaults.

Screen for cross-hybridisation in silico before synthesis. Run every probe against every other probe and against the background genome or transcriptome. Reject or redesign any pair with more than about 12 contiguous complementary bases. This costs an afternoon of compute and saves a resynthesis.

Decouple specificity from pulldown. If the same physical step provides both target recognition and separation from the mixture (as when biotinylated probes are pulled by beads), adding recognition elements adds separation surface and therefore background. If recognition is by a soluble catcher strand and separation is a phase transition of the polymer carrying it, capacity and specificity scale together without adding surface. This is the core design choice in LASSO.

Make the panel redesignable. Prefer architectures in which each target's probe is an independent oligo that can be swapped without touching the rest. Catcher strand libraries are assembled from individual oligos, so a new target is one new oligo and a panel revision is days.

Validate per target, not pooled. After a capture, measure recovery for every target individually (qPCR or per-target read depth), and set an acceptance floor. A pooled number hides dropouts. Iterate on the failures.

Balance abundances. For targets that differ by orders of magnitude in input abundance (a housekeeping transcript versus a rare fusion), either adjust probe stoichiometry or accept that uniform capture will not give uniform depth and plan the sequencing accordingly.

Where multiplexed capture matters most

• Infectious disease panels: dozens of pathogens with divergent sequences, where a fixed commercial panel misses emerging variants and a redesignable one does not.
• Liquid biopsy: a fixed set of cancer genes captured from sub-nanogram cfDNA, where every fragment counts and background costs sensitivity.
• Custom NGS enrichment: a lab's own gene set, iterated as the project evolves.
• rRNA depletion: itself a multiplexed capture of many rRNA fragments, which is why riboLASSO uses a 195-strand catcher library derived from the Adiconis et al. (Nature Methods, 2013) tiling rather than a handful of probes.

How LASSO handles it

Catcher strands are designed to a common Tm, screened in silico for cross-hybridisation, and synthesised individually. They are attached to a soluble polymer; hybridisation to targets triggers a phase transition that pellets the network. Because there is no bead surface, background does not rise as the panel grows, and capacity scales with polymer amount (about 8 to 20 times a streptavidin bead per milligram). Panel revisions are new oligos, not a new pool. We supply per-target validation data with every custom library (Dynamic Matrices, Angewandte Chemie, 2025).