Start a conversation

Platform architecture

From tumor data
to testable mRNA candidates.

MendCipher is being built as a traceable, closed-loop research system—not an automated clinical decision-maker.

The system

Four stages.
One learning loop.

Each stage has an explicit input, method, output, and expert checkpoint. The point is not automation for its own sake; it is disciplined iteration.

01

Profile the tumor

Build a patient-level view without erasing the differences that matter.

Input
Genomic, transcriptomic, and relevant clinical context
Method
Structured integration and quality-aware representation
Output
A contextualized molecular profile
Expert checkpoint
Data quality and biological relevance
02

Interpret the evidence

Connect patient-specific signals to the evidence around targets and mechanisms.

Input
Molecular profile, annotations, publications, and prior evidence
Method
Domain-informed language models with biological constraints
Output
Ranked, traceable therapeutic hypotheses
Expert checkpoint
Evidence strength, uncertainty, and clinical context
03

Design candidates

Move from a prioritized hypothesis to constructs that can be tested.

Input
Selected targets and design objectives
Method
Sequence exploration under expression and manufacturability constraints
Output
Candidate mRNA construct designs
Expert checkpoint
Biological intent, feasibility, and risk
04

Validate and learn

Let experiments challenge the computational view—and improve the next cycle.

Input
Candidate constructs and experimental protocols
Method
Laboratory and preclinical evaluation
Output
Measured evidence and updated design priors
Expert checkpoint
Reproducibility, safety signals, and next-step decisions

Where AI fits

Intelligence narrows the search.
Evidence earns the decision.

AI CAN SUPPORT
  • Connecting heterogeneous biomedical evidence
  • Making hypotheses and provenance easier to inspect
  • Ranking options within explicit constraints
  • Learning from structured experimental feedback
AI DOES NOT REPLACE
  • Expert scientific and clinical judgment
  • Biological and translational constraints
  • Reproducible laboratory validation
  • Regulatory review or treatment decisions

System requirements

Designed to show its work.

01

Traceable

Recommendations should point back to their evidence, assumptions, and constraints.

02

Uncertainty-aware

Confidence is not certainty. Open questions stay visible at every consequential step.

03

Human-accountable

Domain experts review the decisions that shape experimental work.

04

Experiment-led

Computational elegance means little until measured biology supports it.

COLLABORATION / RESEARCH / TRANSLATION

Building at the boundary
of code and biology?

Talk with us