Probabilistic Arithmetic Automata and their Applications

We present probabilistic arithmetic automata (PAAs), a general model to describe chains of operations whose operands depend on chance, along with two different algorithms to exactly calculate the distribution of the results obtained by such probabilistic calculations. PAAs provide a unifying framework to approach many problems arising in computational biology and elsewhere. Here, we present five different applications, namely (1) pattern matching statistics on random texts, including the computation of the distribution of occurrence counts, waiting time and clump size under HMM background models; (2) exact analysis of window-based pattern matching algorithms; (3) sensitivity of filtration seeds used to detect candidate sequence alignments; (4) length and mass statistics of peptide fragments resulting from enzymatic cleavage reactions; and (5) read length statistics of 454 sequencing reads. The diversity of these applications indicates the flexibility and unifying character of the presented framework. While the construction of a PAA depends on the particular application, we single out a frequently applicable construction method for pattern statistics: We introduce deterministic arithmetic automata (DAAs) to model deterministic calculations on sequences, and demonstrate how to construct a PAA from a given DAA and a finite-memory random text model. We show how to transform a finite automaton into a DAA and then into the corresponding PAA.

💡 Research Summary

The paper introduces Probabilistic Arithmetic Automata (PAAs), a formalism that extends finite automata with arithmetic operations whose operands are generated by stochastic processes. A PAA consists of a finite set of states, a transition function that depends on both the current state and a randomly drawn input symbol (according to a finite‑memory text model such as a Markov chain or HMM), and an arithmetic function that updates a numeric accumulator during each transition. The authors provide two exact algorithms for computing the full probability distribution of the accumulator after processing an input of length n. The first algorithm is a dynamic‑programming approach that iteratively updates the joint distribution over state‑value pairs using the transition matrix; its time complexity is linear in the product of the number of states, the size of the value domain, and the input length. The second algorithm exploits the Fourier transform of probability generating functions, allowing convolution of value updates in the frequency domain and reducing the dependence on the value range to a logarithmic factor. Both methods guarantee exact results, unlike Monte‑Carlo or approximation techniques.

A central contribution is a systematic construction pipeline: deterministic arithmetic automata (DAAs) are first built to model deterministic calculations on a concrete sequence, then combined with a stochastic text model by forming the Cartesian product of the DAA’s state space and the text model’s state space. This DAA→PAA transformation preserves the Markov property and yields a PAA whose transition probabilities are inherited from the text model while the arithmetic updates follow the DAA’s deterministic logic. The paper shows how any finite automaton can be turned into a DAA, thereby making the framework applicable to a wide range of problems.

Five diverse applications illustrate the flexibility of PAAs. (1) Pattern‑matching statistics under HMM backgrounds: the authors compute exact distributions for pattern occurrence counts, waiting times, and clump sizes, surpassing earlier methods that only provided expectations or approximations. (2) Exact analysis of window‑based pattern‑matching algorithms such as Shift‑Or and BNDM: by modeling the algorithm’s scanning process as a DAA, they obtain the full cost distribution (number of character comparisons) for arbitrary window sizes and alphabet characteristics. (3) Sensitivity of filtration seeds used in sequence alignment: PAAs quantify the probability that a given seed (including spaced or vector seeds) captures a true alignment, enabling principled seed design. (4) Length and mass statistics of peptide fragments generated by enzymatic cleavage: the cleavage rules are encoded in a DAA, the protein sequence is modeled by an HMM, and the resulting PAA yields exact fragment length and mass distributions, improving peptide‑identification pipelines. (5) Read‑length distribution of 454 pyrosequencing reads: stochastic models of insertion, deletion, and premature termination are combined with a DAA representing the sequencing chemistry, producing accurate predictions of read‑length profiles for experimental planning.

In each case the authors construct the appropriate DAA, apply the DAA→PAA conversion, and then use either the DP‑based or Fourier‑based algorithm to compute the desired distribution. Empirical evaluations demonstrate that the exact methods are computationally feasible for realistic model sizes and provide markedly higher accuracy than existing heuristic approaches.

Overall, the work establishes PAAs as a unifying theoretical tool for exact probabilistic analysis of algorithms and biological processes that involve sequential arithmetic under uncertainty. By separating deterministic computation (captured by DAAs) from stochastic input generation, the framework offers a modular, reusable approach that can be extended to many other domains where probabilistic arithmetic is central.