We compare solutions of puzzles generated from the same protein sequences present in the vaccine formulas SARS 2003 coronavirus. That’s how we created the game. When you play AnalysisMode, you compare 3 or more protein sequences, represented by symbols. Each gameplay helps our artificial intelligence (AI) to find patterns. With enough players we can generate a list of possible formulas, then submit them to our AI that will predict a vaccine formula for SARS-CoV-2. Then we will send it to a laboratory for validation and production.
Indeed, the first puzzle levels are only an introduction to teach our users how to play. When you play further, real difficulty comes and makes humans outperform computers. Meanwhile, any level is already useful as there are 10^24 combinations for this puzzle and each player helps us solve a different one.
Symbols represent different chemical properties of amino acids: polarity (triangle), electrical charge (plus, minus or equal sign), hydrophilicity (water drop), presence of an aromatic ring (hexagon) and ionizability (circle).
Each gameplay will join a pool that is inputted in our created AI to predict a vaccine formula for SARS-CoV-2. Each gameplay brings us closer to the goal of having a vaccine and stopping the pandemic. When players solve puzzles, they are detecting a pattern in an existing vaccine formula that can provide information for the creation of a new vaccine.
In the puzzle, each symbol represents a chemical property of amino acids. When you find a repeating pattern, you are informing the AI about which chemical properties and which amino acids should be together in a vaccine formula. When you play, you will likely get a puzzle combination that no one ever got, because the amount of different puzzle combinations is equivalent to the amount of grains of sand on Earth (10^24).
We are developing a scoring system to avoid misuse of the game. The artificial intelligence behind AnalysisMode is going to be used to predict vaccine formulas only after it learns from hundreds thousands or millions of games. Even if each puzzle solution is not perfect, on average the AI is learning the right patterns.
The game is in prototype phase and, currently, can be solved much faster by a computer than a human. Nonetheless, the results contributed are already describing patterns that can already be used for vaccine formula prediction. Most importantly, by playing you are helping us to validate and develop our method and help us lay the path for a new way of engaging the public in scientific research.
Vaccines take on average 12 years to reach the market. If we want to quickly respond to a crisis such as SARS-CoV-2, it is essential to understand why it takes so long and what we can do to speed up this process. Vaccine development is divided into two stages: vaccine discovery and clinical development. Before being released to the market, a vaccine must go through the first three stages of clinical trials. During this stage, the vaccine is tested in increasingly larger numbers of people and researchers meticulously study if the vaccine is safe and effective. Many factors must be considered such as different age groups and pre-existing health conditions. Also, researchers must determine how to keep the vaccine safe and effective by developing rigorous packing and storage standards. Clinical development normally takes six to eight years, however, trying to reduce this time comes at considerable risk. For this reason, we must focus on what we can do. AnalysisMode can shorten the exploratory stage of vaccines from a few years to a few weeks by joining the innate pattern-recognition powers of the human brain and artificial intelligence.
The model behind AnalysisMode answers the question: given the individual biochemical properties of each amino acid in a sequence is the sequence likely to be a B-cell epitope? This is possible by analyzing the experimentally validated epitopes of other known viruses. Other similar models developed so far mostly rely on a single chemical property of amino acids: how well they bind to water. This approach is very useful for many virus families but not for coronaviruses. Therefore, our tool considers four additional chemical properties: polarity, electrical charge, ionizability and presence of an aromatic ring. The sequences predicted by the model can then be sent to the lab for experimental validation and production
While computers are absolutely unbeatable in performing mathematical calculations, the human brain can easily outperform the world’s fastest computer in many tasks. In particular, we are very good at recognizing visual patterns. Machine learning is a branch of computer science that deals with pattern recognition and one way by which machines can “learn” to recognize patterns is by studying the results of tasks performed by humans. In AnalysisMode humans can play a simple game to map complex associations between the biochemical properties between amino acids within an epitope. Each time you play the game, you are creating an example from which the AI can learn.
In the current version of AnalysisMode the game features a relatively simple task. Future development of the tool will focus on improving the game design to harness the human brain's natural ability to detect patterns while also providing an enjoyable experience for the user.
We are randomly selecting protein sequences and comparing the biochemical properties of the amino acids.
In each level, the player compares a set of 3 to 5 sequences of amino acids (represented by letters), one by one. For each column, the player must find the symbol appearing most often.
I guess it depends on how fast you code :). Thanks to the hard work of hundreds of scientists there are more than 40 vaccines for SARS-CoV-2 currently in preclinical or clinical trials. So, hopefully, the vaccine candidates predicted by our method will arrive too late. The SARS-CoV-2 crisis inspired us, but the method itself can be used to discover other vaccines and also pharmaceutical drugs.