Logical design and logical operation of the molecular computer. a, Function and modular organization of the molecular computer. b, Example diagnostic rules for simplified models of SCLC and prostate cancer, indicating overexpression ( ↑ ) or underexpression ( ↓ ) of a disease-related gene. The first rule states that if genes ASCL1, GRIA2, INSM1 and PTTG1 are overexpressed then administer the ssDNA molecule TCTCCCAGCGTGCGCCAT (oblimersen), purported to be an antisense therapy drug for SCLC. The second rule states that if the genes PPAP2B and GSTP1 are underexpressed and the genes PIM1 and HPN are overexpressed then administer the ssDNA molecule GTTGGTATTGCACAT, purported to be a drug for prostate cancer. c, Transition diagram of the diagnostic automaton. d, The computation that diagnoses prostate cancer.

Operation of the molecular computer. The complete sequences for all molecules shown are given in the Supplementary Methods. a, Part of the computation path for the diagnostic molecule for prostate cancer with all molecular indicators present, ending in drug release. The initial diagnostic molecule consists of a diagnosis moiety (grey) that encodes the left-hand side of the diagnostic rule () and a drug-administration moiety (light purple) incorporating an inactive drug loop (dark purple). At each computation step, the prevailing transition is shown, except for the processing of the symbol PIM1 ↑ , for which details of the stochastic choice, accomplished by a regulated pair of competing transition molecules, are shown (dashed box, see c). b, Regulation of the two transitions for PIM1 ↑ by sub-sequences (tags) of overexpressed PIM1 mRNA, resulting in a relatively high level of the Yes Yes transition molecules and low level of the Yes No molecules. Each transition molecule contains regulation (green, orange) and computation (blue, grey) fragments. The ‘inactivation tag’ of PIM1 mRNA (light orange) displaces the 5′ → 3′ strand of the transition molecule Yes No and destroys its computation fragment. The ‘activation tag’ of PIM1 mRNA (light green) activates the transition molecule Yes Yes. Initially, a protecting oligonucleotide (green) partially hybridizes to the 3′ → 5′ strand of the transition molecule and blocks the correct annealing of its 5′ → 3′ strand. The ‘activation tag’ displaces the protecting strand, allowing such annealing and rendering an active Yes Yes transition. Ideally, one PIM1 mRNA molecule inactivates one Yes No and activates one Yes Yes transition molecule. c, Stochastic processing of the symbol PIM1 ↑ by a regulated pair of competing transition molecules. The probability of a Yes → Yes transition is high, resulting in a high level of diagnostic molecules in the state Yes and a low level in state No. d, Combining computation results for both types of diagnostic molecules, both with high Yes and low No final states, results in high release of drug and low release of drug suppressor, and hence in the administration of the drug.

Experimental demonstration of diagnosis. a, Regulation of competing transitions by pTRI-Xef mRNA (bar chart) representing an example molecular indicator. The gel shows positive (blue and pink) and negative (green) transition, each in both active and inactive states, the relative concentration of which is regulated by mRNA concentration. F, carboxyfluorescein; R, tetramethyl rhodamine; Y, Cy5. b, Validation of the diagnostic automata with the diagnostic rules for SCLC and prostate cancer (PC). Each lane shows the result of a diagnostic computation for the indicated combination of molecular indicators. The location of the initial, intermediate and output molecules is indicated on the left and a predicted trace of the diagnostic computations is shown on both sides. Some, but not all, intermediates are visible owing to their incomplete processing by the automaton. c, Parallel diagnosis of two disease models by two diagnostic automata. The diagnosed disease model contains the two indicators of SCLC checked by the diagnostic string PTTG1 ↑ CDKN2A ↑ and the two indicators of prostate cancer checked by the string PIM1 ↑ HPN ↑ . The gene CDKNA2 is another indicator for SCLC when overexpressed, and it is used here for technical reasons. The presence of indicators for each disease model as well as the expected diagnostic output by each automaton are indicated above the lanes.

Experimental demonstration of drug administration. a, Release of an active drug by a drug-release PPAP2B ↓ GSTP1 ↓ PIM1 ↑ HPN ↑ diagnostic molecule, showing absolute amount of the active drug versus positive diagnosis probability. b, Different diagnostic outcomes are modelled using active transition molecules with a mixture of equal amounts of the drug-release and drug-suppressor release moieties for the diagnostic string PPAP2B ↓ GSTP5 ↓. Each lane shows the distribution of drug-administration moieties, active drug, excess drug suppressor and drug/drug-suppressor hybrid, as indicated. c, Variation in the distribution of active drug, excess drug suppressor and drug/drug-suppressor hybrid for a given diagnostic outcome and for varying relative amounts of drug release and drug-suppressor release diagnostic moiety. y-axis units in b and c are arbitrary units.