As highthroughput biological sequencing becomes faster and cheaper, the need to extract useful information from sequencing becomes ever more paramount, often limited by lowthroughput experimental characterizations. For proteins, accurate prediction of their functions directly from their primary aminoacid sequences has been a long standing challenge. Here, machine learning using artificial recurrent neural networks (RNN) was applied towards classification of protein function directly from primary sequence without sequence alignment, heuristic scoring or feature engineering.The RNN models containing longshorttermmemory (LSTM) units trained on public, annotated datasets from UniProt achieved high performance for inclass prediction of four important protein functions tested, particularly compared to other machine learning algorithms using sequencederived protein features. RNN models were used also for outofclass predictions of phylogenetically distinct protein families with similar functions, including proteins of the CRISPRassociated nuclease, ferritinlike iron storage and cytochrome P450 families. Applying the trained RNN models on the partially unannotated UniRef100 database predicted not only candidates validated by existing annotations but also currently unannotated sequences. Some RNN predictions for the ferritinlike iron sequestering function were experimentally validated, even though their sequences differ significantly from known, characterized proteins and from each other and cannot be easily predicted using popular bioinformatics methods. As sequencing and experimental characterization data increases rapidly, the machinelearning approach based on RNN could be useful for discovery and prediction of homologues for a wide range of protein functions.peer-reviewed)

Genetically encoding the synthesis of functional nanomaterials such as magnetic nanoparticles enables sensitive and non-invasive biological sensing and control. Via directed evolution of the natural iron-sequestering ferritin protein, we discovered key mutations that lead to significantly enhanced cellular magnetism, resulting in increased physical attraction of ferritin-expressing cells to magnets and increased contrast for cellular magnetic resonance imaging (MRI). The magnetic mutants further demonstrate increased iron biomineralization measured by a novel fluorescent genetic sensor for intracellular free iron. In addition, we engineered Escherichia coli cells with multiple genomic knockouts to increase cellular accumulation of various metals. Lastly to explore further protein candidates for biomagnetism, we characterized members of the DUF892 family using the iron sensor and magnetic columns, confirming their intracellular iron sequestration that results in increased cellular magnetization.Inorganic nanomaterials have been used in a wide range of biological applications including fluorescent or plasmonic labelling for imaging, magnetic labelling for extraction and high throughput sequencing and drug-delivery 1-3 . However, unlike genetically-encoded labels such as green fluorescent protein (GFP), chemically synthesized inorganic nanomaterials, despite their versatile physical and chemical properties, are ultimately limited in their biological application by their lack of integration with the genetic circuitry of the cell. Synthetic biology can bridge this gap by programming cells to controllably synthesize their own nanomaterials in response to biological signals. Those nanomaterials can be further tailored within cells to interact with other components and transduce biological signals downstream.There are few examples of bio-synthesized inorganic nanomaterials in Nature. Certain species of bacteria and archaea can mineralize nanoparticles via proteins or metabolites that reduce toxic metal cations 1,4 . Notably, magnetotactic bacteria of the genus Magnetospirillum naturally synthesize crystalline magnetite nanoparticles and align them as a passive navigation compass for the cell in its natural environment [5][6][7] . Despite speculation on the presence of similar inorganic magnetic nanoparticles in animals such as fish and humans, no such biomineralization pathways have been confirmed so far [8][9][10][11] . However, all cells do use inorganic bio-mineralization to maintain near constant concentrations of essential trace metals via high affinity chelators and storage proteins for times of excess. One prominent example are the ferritins, a ubiquitous class of proteins found in all domains of life that play a crucial role in iron homeostasis 2,[12][13][14][15][16][17][18] . Ferritins form shells composed of 24 monomers each, creating an inner cavity in order to store iron in a hydrated amorphous form of iron oxide similar to the mineral ferrihydrite. (Figure 1a,b) Iron oxide is biocompatible and magnetic depe...