Detecting and Recovering from Faults in Programmed Molecular Systems

Abstract:

For reliable operation of programmed molecular systems, the capability to detect faults when they occur and to initiate recovery is essential. Since a faulty system cannot be depended upon to report its own failure, a monitoring device is needed. The work presented here describes the design of one such fault protection device, called a molecular watchdog timer. A watchdog timer is a standard device used in safety-critical applications to monitor the health of a system and issue an alarm (“bark”) if the monitored system fails. Often, as here, the watchdog timer monitors a system’s health by watching for periodic “heartbeats” from it. If a heartbeat is not received for a set time period, the monitored system is assumed to have failed, and the watchdog timer issues an alarm and/or triggers appropriate recovery actions. Thus, in the design of our molecular watchdog timer, the absence of a heartbeat for a preset time can produce, with arbitrarily high likelihood, a cascading reaction that generates a fluorescent signal visible with FRET. The molecular watchdog timer itself is comprised of three components: an absence detector, a threshold filter, and a signal amplifier. These components are separately designed as chemical reaction networks. They are verified both individually and when composed using model-based simulation, probabilistic model checking and formal analysis. To verify its embeddability, the molecular watchdog timer is assigned to monitor the health of a clock (based on the three-species stochastic Lotka-Volterra oscillator) that outputs a periodic heartbeat. Another recovery solution presented here is more demanding but also more appropriate for eventual use in biocompatible environments because it does not require operator intervention. Here, when the molecular watchdog detects the failure of the clock, it handles recovery autonomously by having a repair component reboot the clock. This design is shown to be realizable at scale. If the “ladders” used as “delays” have size logarithmic in the number of molecules, then the expected time to a false alarm can be as large as desired. A Chernoff bound permits the guarantee that, with arbitrarily high probability, the actual time to a false alarm is nearly as great as the expected time. Moreover, if the heartbeat stops, the alarm species is produced very quickly, so the molecular watchdog timer’s alarm will be triggered. The molecular watchdog timer offers a simple yet powerful means to prevent faults in cybermolecular systems from leading to interruptions in their reliable operation. This work is supported in part by by NSF grants 1247051 and 1545028. An earlier version was presented at DNA21.