Fig. 1: Framework of the DNA-encoded computational decision tree.

a A three‑node decision tree with three input features (Weather, Distance, and Work) yielding six edges and two leaf nodes (“Go” and “Not Go”). The right panel shows how this mini-sized tree can be embedded within a larger multi-layer tree. b Schematic of a node-encoding molecule. Each molecule comprises four sequence domains: Domain 1 (parent node), Domain 2 (current node), Domain 3 (edge identifier), and Domain 4 (child node). Domains are color‑ and letter‑denoted, with arrowheads marking 3′ ends. c A equivalent dual-rail Boolean logic circuit, comprising 18 gates (six NOT, six AND, six NAND) to implement the rule set embedded within the decision tree shown in (a). d Untraversed tree and a one-hop traversal from node A to node C. Right: molecular implementation of the one-hop traversal from A to C through an entropy-driven strand displacement reaction cascade. The root node is activated in its initiate state, allowing its input to directly displace an output strand encoding the activator for the child node. For example, the input of node A can release an activator that first displaces the blocker on node C, exposing the hidden toehold and activating the node C. Only after this activation step can the node-specific input hybridize with the toehold and release the activator of the next child-node while reproducing the parent node-derived activator. Traversing a node requires prior activation, with “untraversed,” “activated,” and “traversed” denoting successive states of the same node.