Innocent until proven guilty: Privacy-preserving search over a central CODIS criminal database from the field

The presumption of innocence (i.e., the principle that one is considered innocent until proven guilty) is a cornerstone of the criminal justice system in many countries, including the United States. DNA analysis is an important tool for criminal investigations1. In the U.S. alone, it has already aided in over half a million investigations using the Combined DNA Index System (CODIS) and associated DNA databases2. CODIS includes DNA profiles of crime scene forensic samples, convicted offenders, missing persons and more. The CODIS framework is currently used by over 50 other countries3 including much of Europe, Canada, China and more. During investigations, DNA samples can be collected from multiple individuals who may have had access to, or were found near a crime scene, in the hope of finding a single criminal match4. Controversially, CODIS samples are sometimes retained from adults and juveniles despite not yielding any database match4–6. Here we introduce a cryptographic algorithm that finds any and all matches of a person’s DNA profile against a CODIS database without revealing anything about the person’s profile to the database provider. With our protocol, matches are immediately identified as before; however, individuals who do not match anything in the database retain their full privacy. Our novel algorithm runs in 40 seconds on a CODIS database of 1,000,000 entries, enabling its use to privately screen potentially-innocent suspects even in the field.

Predicting the Function of Hypothetical Genes in Genomes of Bioleaching Microorganisms

A particularly challenging problem in genome annotation is to attribute function to genes annotated as “hypothetical, no known function”. These typically account for about 40% of all genes regardless of the genome. Some of these are “orphan” genes and are not found in any other genome. Some of these could encode species specific proteins and so are particularly interesting for evaluating novel metabolic potential and for understanding the evolution of genes and genomes. Several similarity and non-similarity bioinformatics tools exist that help predict function of hypotheticals, but none are able to suggest function for more than a few percent and the annotation of the others remains a formidable task. We have developed a bioinformatics tool called AlterORF (www.AlterORF.cl) that is able to identify alternate open reading frames (ORFs) embedded within annotated genes. Analysis of over 2 million genes in over 700 completely sequenced genomes reveals that alternate ORFs of substantial length (potentially encoding 70 amino acids or more) are surprisingly common, especially in G+C rich genomes. During our examination of these alternate ORFs, we uncovered hundreds of examples where the alternate ORF has a significant hit with databases of motifs and domains (e.g. CDD, Pfam) and where the actual annotated gene is described as hypothetical and has no database match. This strongly suggests that the annotated gene has been incorrectly identified and that the alternate ORF is the real gene. We describe the evaluation of the following genomes of bioleaching microorganisms and others that reside in similar ecological niches using AlterORF: Acidithiobacillus ferrooxidans (2 strains), Leptospirillum type II, Methylacidiphilum infernorum, Picrophilus torridus, Sulfolobus acidocaldarius, S. solfataricus, S. tokodaii, Thermodesulfovibrio yellowstonii, Thermoplasma acidophilum and T. volcanium. Examples of novel genes from these microorganisms and their suggested roles in metabolism will be described.

Landscape Image Retrieval with Query by Sketch and Icon

This paper reports a new method for retrieving landscape images using a sketch and icons as a query. Based on the proposal, first, a user sketches lines expressing contours of landscape elements such as mountains and forests and attaches icons expressing landscape elements to the sketch. Second, whether individual images in a database match with the layout expressed by the sketch and icons is judged with principal component analysis and pattern recognition. From experimental results, we have confirmed that the proportion of the correct images ranked within top 10 of retrieval results is 80% in an average.